Page 1 of 1 - core.ac.uk · Elektron spin resonantie (ESR) is hiervoor de uitgelezen techniek....

of 1

13/06/2007http://www.kuleuven.be/huisstijl/downloads/sedes.jpg

Katholieke Universiteit LeuvenFaculteit WetenschappenDepartement Natuurkunde en SterrenkundeLaboratorium voor Halfgeleiderfysica

Electron spin resonance probing of point

defects in functional oxides and Si/high-κ

structures

Katrijn Clemer

Promotor:Prof. Dr. A. Stesmans

Proefschrift ingediend tothet behalen van de graad vanDoctor in de Wetenschappen

2007

Proefschrift voorgedragen tot het behalen van het doctoraatin de Wetenschappen doorKatrijn Clemer

En daarvoor publiekelijk verdedigdop 25 oktober 2007 om 17 u in auditorium 05.11,Celestijnenlaan 200 D3001 HeverleeBelgie

De examencommissie is samengesteld uit:

PromotorProf. Dr. A. Stesmans

LedenProf. Dr. G. AdriaenssensProf. Dr. V.V. Afanas’evProf. Dr. M. Van der AuweraerProf. Dr. C. GlorieuxDr. M. Houssa

Acknowledgments

Dank aan allen die mede verantwoordelijk gesteld kunnen worden voor hettot stand komen en volbrengen van deze thesis.

Voor de talrijke wetenschappelijke discussies en begeleiding dank ik de pro-motor van dit werk Prof. Dr. A. Stesmans, Prof. Dr. V. V. Afanas’ev,Pieter en alle andere leden van de afdeling Halfgeleiderfysica.

Voor het kritisch lezen en becommentariseren van dit werk dank ik de ledenvan de jury: Prof. Dr. A. Stesmans, Prof. Dr. V. V. Afanas’ev, Prof. Dr.G. Adriaenssens, Prof. Dr. M. Van der Auweraer, Prof. Dr. C. Glorieux,en Dr. M. Houssa.

Voor de persoonlijke ondersteuning in goede en in moeilijke tijden dank ik,Pieter, Sheron, Koen, Maria, mijn familie, en in het bijzonder Marc.

Follow the yellow brick road...–L. F. Baum–

Contents

List of abbrevations 12

List of symbols 13

Samenvatting 15

Introduction 19

1 Electron spin resonance probing of defects 231.1 Electron spin resonance: theory and experiment . . . . . . 23

1.1.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . 231.1.2 Practice . . . . . . . . . . . . . . . . . . . . . . . . 31

1.2 Defects in the Si/SiO2 system . . . . . . . . . . . . . . . . 351.2.1 Pb-type interface defects . . . . . . . . . . . . . . . 371.2.2 Intrinsic defects in a-SiO2 . . . . . . . . . . . . . . . 411.2.3 Extrinsic defects in a-SiO2 . . . . . . . . . . . . . . . 47

1.3 Defects in the Si/high-κ system . . . . . . . . . . . . . . . 521.3.1 The Si/high-κ interface . . . . . . . . . . . . . . . . 541.3.2 Intrinsic defects in high-κ oxides . . . . . . . . . . . 561.3.3 Impurity related defects in high-κ oxides . . . . . . 60

2 Fundamental point defects in nm-sized silica particles: prob-ing of the network structure 632.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 632.2 Former studies . . . . . . . . . . . . . . . . . . . . . . . . . 652.3 Interactions with SiO at elevated temperatures . . . . . . . 702.4 Experimental details . . . . . . . . . . . . . . . . . . . . . . 71

2.4.1 Samples . . . . . . . . . . . . . . . . . . . . . . . . 712.4.2 ESR spectrometry . . . . . . . . . . . . . . . . . . . 71

2.5 ESR probing of occurring point defects . . . . . . . . . . . 73

9

10

2.5.1 The E′-type center . . . . . . . . . . . . . . . . . . 742.5.2 A newly isolated ESR signal . . . . . . . . . . . . . 1002.5.3 The OHC type centers . . . . . . . . . . . . . . . . 1042.5.4 The EX center . . . . . . . . . . . . . . . . . . . . 1082.5.5 The methyl radical . . . . . . . . . . . . . . . . . . 108

2.6 Summary and Conclusions . . . . . . . . . . . . . . . . . . 113

3 P-associated defects in the high-κ HfO2 and ZrO2 insulatorsrevealed by ESR 1173.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 1173.2 Phosphorus diffusion . . . . . . . . . . . . . . . . . . . . . 1183.3 Experimental details . . . . . . . . . . . . . . . . . . . . . . 121


3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1223.5 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

3.5.1 Spectra simulations . . . . . . . . . . . . . . . . . . 1283.5.2 Structural model . . . . . . . . . . . . . . . . . . . . 1293.5.3 LCAO analysis . . . . . . . . . . . . . . . . . . . . . 1293.5.4 Defect densities . . . . . . . . . . . . . . . . . . . . 131

3.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 1323.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

4 Paramagnetic point defects in (100)Si/LaAlO3 structures:nature and stability of the interface 1394.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 1394.2 The (100)Si/LaAlO3 structure . . . . . . . . . . . . . . . . 1404.3 Experimental details . . . . . . . . . . . . . . . . . . . . . . 143


4.4 Experimental results and analysis . . . . . . . . . . . . . . 1444.4.1 Observed defects: Pb’s and EX . . . . . . . . . . . 1444.4.2 Defect densities vs. Tan . . . . . . . . . . . . . . . . 148

4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 1504.5.1 Interlayer formation and disintegration . . . . . . . 1504.5.2 Influence of oxide crystallization and interdiffusion . 1504.5.3 Absence of Pb1 defects . . . . . . . . . . . . . . . . 151

4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

11

5 Origin of optical absorption in the range 4.8-4.9 eV and5.7-5.9 eV in silica implanted with Si or O. 1555.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 1555.2 Experimental details . . . . . . . . . . . . . . . . . . . . . . 158

5.2.1 Samples and ion implantation . . . . . . . . . . . . 1585.2.2 ESR spectrometry . . . . . . . . . . . . . . . . . . . 158

5.3 ESR results . . . . . . . . . . . . . . . . . . . . . . . . . . . 1605.3.1 Oxygen implanted samples . . . . . . . . . . . . . . 1605.3.2 Silicon implanted samples . . . . . . . . . . . . . . . 163

5.4 Analysis and discussion . . . . . . . . . . . . . . . . . . . . 1675.4.1 The 4.8-4.9 eV optical absorption band . . . . . . . 1675.4.2 The 5.7-5.9 eV optical absorption band . . . . . . . 170

5.5 Summary and conclusions . . . . . . . . . . . . . . . . . . . 173

Summary and conclusions 177

References 181

List of Publications 199

12

List of abbreviations

ALCVD atomic layer chemical vapor depositionBET Brunauer-Emmet-TellerBHODC bridged hole-trapping oxygen deficiency centerBPSG borophosphosilicateCVD chemical vapor depositioncw continuous waveDFT density functional theoryEOT equivalent oxide thicknessESR electron spin resonanceFESEM field emission scanning electron microscopyFTIR fourier transform infrared absorptionFWHM full width at half amplitudehf hyperfineHRTEM high-resolution transmission electron microscopyIC integrated circuitITRS international technology roadmap for semiconductorsLCAO linear combination of atomic orbitalsMBD molecular beam depositionMD molecular dynamicsMEIS medium-energy ion scatteringMOCVD metallo-organic chemical vapor depositionMOS metal-oxide-semiconductorMOSFET metal-oxide-semiconductor field-effect transistorNBOHC non-bridging oxygen hole centerNCVD ALCVD using the nitro precursor Hf(NO3)4

NMR nuclear magnetic resonanceOHC oxygen-related hole centerPDA post-deposition annealingPL photoluminescencePOHC phosphorus oxygen hole centerPOR peroxy-radicalPSG phosphosilicateRTA rapid thermal annealingSIMS secondary ion mass spectroscopyVUV vacuum ultra violetXPS x-ray photoelectron spectroscopy

13

List of symbols

α2 s character of the orbital composition of the unpaired electronαSiO2 thermal expansion coefficient of quartzβ2 p character of the orbital composition of the unpaired electron∆γ g matrix anisotropyη2 localization of the electron wave function on an atomθ O–Si–defect bond angleκ dielectric constantκls line shape factor~µ magnetic momentµB Bohr magneton~µe intrinsic angular momentum~µL orbital angular momentumν microwave frequencyτc correlation timeχ magnetic susceptibility

A hyperfine tensorA‖, A⊥ parallel and perpendicular components of an axial AAiso, b, c isotropic, anisotropic, and orthorhombic part of AAs, Ap atomic hyperfine coupling constantsApp peak-to-peak hight~B magnetic field inductionBm modulation amplitude∆Bpp peak-to-peak line widthdox grown oxide thicknessg, g spectroscopic splitting factor/matrixg‖, g⊥ parallel and perpendicular components of an axial gg0 free electron g factorgc zero crossing g value, central g valuegav average g valueh Planck’s constantHS spin HamiltonianHez electronic Zeeman HamiltonianHD spin-spin HamiltonianHhf hyperfine Hamiltonian

14

Hnz nuclear Zeeman HamiltonianI signal intensity~J total angular momentum~L orbital angular momentumMI nuclear spin quantum numberMJ magnetic quantum numberNS number of spinsPµ incident microwave power~S electron spin angular momentumT measurement temperatureTan (observation) temperatureTC Curie temperature

Samenvatting

Tegenwoordig verwachten we dat onze gadgets gebaseerd op halfgeleidertech-nologie steeds kleiner, sneller, multifunctioneler, en –als het even kan– ooknog goedkoper worden. Gedurende jaren is de industrie er in geslaagd omtegemoet te komen aan deze eisen door onder andere miniatuur GSMs,MP3-spelers, foto- en videocamera’s, en labtops op de markt brengen. Hetwoord ’nano’ is zelfs boven zijn wetenschappelijke betekenis (eenheid pre-fix) uitgestegen en werd het toverwoord in de reclame wereld om alles, vanelektronica tot waspoeder, aan de man te brengen.

De elementaire eenheid die decennia lang aan de basis lag van allegeavanceerde toepassingen was de Si/SiO2-structuur. Het succes van dehalfgeleiderindustrie was tot nu toe gebaseerd op de schaalbaarheid van deSi/SiO2-gebaseerde micro-elektronica: Krachtigere chips konden bekomenworden door de afmetingen van de metaal-oxide-halfgeleider (MOS) com-ponenten steeds kleiner te maken. Op dit moment is het punt echter bereiktwaar verdere ’schaling’ van de Si/SiO2-gebaseerde componenten onmogelijkwordt vermits dit zou leiden tot zware problemen met de performantie enbetrouwbaarheid van de MOS componenten. De nieuwe generatie transi-storen zal alternatieve ’nieuwe’ isolerende poort materialen nodig hebben.Een voor de hand liggende oplossing is het gebruik van een isolator met eenhogere dilektrische constante (κ) waardoor een fysisch dikkere oxide laagkan gebruikt worden zonder dat de capaciteit van de poort hierbij moetinboeten.

Het gebruik van zulk een alternatief poort oxide in de plaats van het uit-gebreid bestudeerde SiO2 oxide brengt natuurlijk onverwachte problemenmet zich mee zoals de vorming van een tussenlaag en diffusie van addi-tieven. Er zijn al talrijke studies die handelen over verschillende van dezekandidaat hoge-κ materialen maar uiteindelijk blijven we achter met meervragen dan antwoorden. In een poging om enkele van deze vragen te beant-woorden werden in deze thesis drie van deze hoge-κ materialen bestudeerd:ZrO2, HfO2, en LaAlO3.

15

16 Samenvatting

Een ander gevolg van de huidige nano-rage is de stijgende populariteitvan, o.a., nano-deeltjes, nano-draden, en nano-buisjes. Wanneer men oxidesschaalt tot deze kleine dimensies (∼10−8-10−9 m) veranderen echter hunelektrische en optische eigenschappen, waadoor onderzoekers weer van vooraf aan kunnen beginnen met het bepalen van de materiaaleigenschappen.

Verwikkeld in een strijd om oplossingen te vinden voor al deze nieuweproblemen die zich aandienen is zowel het toegepaste als het fundamenteleonderzoek naar nano-structuren de laatste jaren exponentieel gegroeid.Hierbij werd de toenemende impact van de aanwezigheid van defectenblootgelegd, zowel in de kern van de nano-materialen als aan de halfgelei-der/oxide grenslaag in MOS structuren. De aanwezigheid van n enkel punt-defect kan een bepalend vaak negatief effect hebben op de optische en elek-trische eigenschappen zoals fotoluminescentie, geleiding, en dopering vannano-deeltjes. Om een beter begrip te krijgen van de negatieve invloedendie deze puntdefecten kunnen hebben is het van cruciaal belang dat wekennis kunnen verwerven aangaande de fysische structuur van die defecten.Elektron spin resonantie (ESR) is hiervoor de uitgelezen techniek. Voor-waarden zijn dat de defecten een magnetisch moment (”spin”) hebben en involdoende aantallen aanwezig zijn in het te onderzoeken systeem (typisch1010-1012 magnetisch centra). In deze thesis zal ESR dan ook aangewendworden om optredende puntdefecten te monitoren, karakteriseren, en iden-tificeren in een aantal materialen die mogelijk gebruikt kunnen worden intoekomstige halgeleider nano-technologien.

In een eerste fase werden pyrolytisch gevormde SiO2 nano-deeltjes bestu-deerd, met de bedoeling om via de analyse van de ESR-eigenschappenvan ingebedde intrinsieke puntdefecten, informatie te verkrijgen op releatomaire schaal van de invloed van de sterke dimensionele reductie. Terverhoging van de detectiviteit en resolutie werd het ESR-onderzoek uitge-voerd in combinatie met verschillende types van bestraling, zoals UV, VUV,60Co-gamma straling. Voor het eerst werden in de nano-deeltjes verschil-lende puntdefecten gedetecteerd zoals E′, het zuurstof peroxy-radicaal, hetmethyl radicaal, en een ongekend signaal met axiale symmetrie (g‖=2.0041,g⊥=2.0027). Een grondige studie van de ESR-parameters van deze defectenin functie van onder andere de bestraling en de postdepositie warmtebehan-deling leverde een bijkomend inzicht in de specifieke structuur van de nano-deeltjes op atomair niveau. Er werd experimenteel bewijs geleverd dat erin deze nano-deeltjes twee verschillende systemen E′ centers aanwezig zijn.De specifieke ESR parameters (bvb. g matrix) van de E′ centers van eeneerste bad werden zeer gelijkend gevonden aan deze van het goed gekende

Samenvatting 17

E′γ center in bulk fused SiO2. De E′ centers behorende tot het tweede badgeven blijk van een verschillende nuldoorgang g waarde en lijnvorm. Dezeverschillen werden toegeschreven aan variaties in de locale structuur. Ditlaatste E′ systeem lijkt zich te bevinden in de buitenste SiO2 lagen vande nano-deeltjes structureel verschillend van bulk SiO2. Verder kon uiteen gedetailleerde studie van de hyperfijn splitsing van de E′ centers in dekern van de nano-deeltjes worden afgeleid dat deze kern dichter is. Dezeverdichting is waarschijnlijk geassocieerd met de aanwezigheid van meerkleinere ringen in het SiO2 netwerk.

Een volgend hoofdstuk handelt over de observatie van P-gerelateerdepuntdefecten in nm-dikke P-gemplanteerde HfO2 lagen op (100)Si onder-worpen aan een postfabricatie thermische behandeling (verhitting in N2-gas voor temperaturen tussen 500 en 900 ◦C) en in ZrO2 poeder. Dehoofd g matrices en hyperfijn tensoren van de geobserveerde defecten wer-den bekomen door de X, K, en Q-band spectra consistent te simuleren. Debekomen resultaten werden vergeleken met de goed gekende P-geassocieerdedefecten in silica. Na deze analyse werden beide geobserveerde defectentoegewezen aan een P2-type defect –een P atoom dat een Hf of Zr atoomvervangt. Beide defecten werden geobserveerd in de monokliene fase vande hoge-κ oxiden met het ongepaarde elektron sterk gelokaliseerd op hetP atoom. Een beduidende fractie van de gencorporeerde P onzuiverhedenresulteerde in een ESR actief P2 defect. Vermits er een versterkte diffusievan additieven (P) werd vastgesteld in HfO2 lagen is de identificatie vanresulterende P-geassocieerde defecten van cruciaal belang. Deze defectenkunnen immers optreden als schadelijke ladingsvallen.

De atomaire natuur van de (100)Si/LaAlO3 grenslaag is het onderwerpvan hoofdstuk vier. De nm-dunne amorfe LaAlO3 lagen hebben een hogedielectrische constante (κ∼20-27). In de oorspronkelijke samples wezen deK-band ESR metingen op de afwezigheid van een Si/SiO2 type grenslaagvermits de archetypische Si/SiO2 grenslaag defecten (Pb1, Pb0) niet werdengeobserveerd. De situatie blijft onveranderd na verdere warmtebehandelin-gen to 800 ◦C. De Si/LaAlO3 grenslaag blijkt thermisch stabiel en abruptin tegenstelling tot andere Si/hoge-κ structuren. Na warmtebehandelingenin het temperatuursgebied 800-860 ◦C begint zich echter een Si/SiO2-typegrenslaag te vormen: Pb0 defecten worden geobserveerd en iets later intemperatuur ook het EX center –een SiO2-geassocieerd defect. Dit laatstewijst op een substantile structurele/compositie modificatie. Het pieken vande defect dichtheden vs. warmtebehandelingstemperaturen curven duidterop dat de SiOx natuur van de grenslaag terug opbreekt na warmtebehan-

18 Samenvatting

delingen bij temperaturen groter dan 930 ◦C. Dit is mogelijk gerelateerdmet kristallisatie en/of silicaat vorming.

In een laatste hoofdstuk werd gekeken naar de invloed van ionenimplan-tatie (O en Si) in amorf SiO2. In samenwerking met R. Weeks (VanderbiltUniversity, USA), R. Magruder (Belmont University, USA) en R. Weller(Vanderbilt University, USA) werden de geobserveerde ESR-defect dichthe-den vergeleken met de optische absorptie banden rond 4.8 eV en 5.3 eV .Zo kon informatie verkregen worden over de bron van de geobserveerdeoptische absorpties.

Introduction

Today, we want our devices based on semiconductor devices to be eversmaller, faster, more multi-functional, and, if possible, also cheaper. Foryears now industry has been able to meet these demands bringing e.g.,tiny cell phones, MP3-players, photo and video cameras, and laptops onthe market. The word ’nano’ has even surpassed its scientific meaning(unit prefix) and became the magic word to sell electronic devices and evenwashing-powder.

Up to present, the success of semiconductor-based industry has been thescalability of the microelectronics. More powerful chips could be realizedthrough scaling of the metal-oxide-semiconductor (MOS) device dimen-sions. Using the Si/SiO2 entity as basic building block, this down-scalingwas successfully accomplished for gate thicknesses down to the 1-nm range.At this very moment, however, semiconductor industry has reached the endof the road for pure Si/SiO2 based technology.

Further down-scaling of the SiO2 gate oxide layer would induce excessiveleakage currents because of direct tunneling, and moreover, serious reliabil-ity problems. This means that to keep on track with the requested advancesin semiconductor devices an alternative ”new” gate insulating material willbe required for future generations of MOS devices. As an obvious solution,an insulator with a higher value of dielectric constant (κ) would allow oneto use a physically thicker oxide layer retaining the same gate capacitance.

The use of such an alternative insulator instead of the extensively studiedSiO2 oxide brings about unforeseen problems associated with, e.g., inter-layer formation and dopant penetration. Many candidate materials havebeen studied, leaving us with more questions than answers. In search forsome fundamental answers three of these so-called high-κ materials, ZrO2,HfO2, and LaAlO3, were studied in this thesis and are the subject of chap-ters 3 and 4.

Another outcome of the current nano-wave is the rising popularity of,e.g., nanoparticles, nanowires, and nanotubes. Scaling down the SiO2 di-

19

20 Introduction

electric to such small dimensions, however, changes its electrical and opticalproperties, opening a whole new area of research. The characterization ofthe structure of SiO2 nanoparticles is the subject of chapter 2.

The last few years, applied and fundamental research of nano-structureshas grown exponentially revealing the upcoming relative impact of defectsin the core of the nano-materials (nanoparticles, nanotubes, ultrathin lay-ers) as well as at the interface. The study of point defects subsequentlybecame of general interest since the presence of one single point defectcan play a crucial definite part in the optical and electronic properties,such as, e.g., photoluminescence, doping of semiconductor nanoparticles,and conduction (in e.g., carbon nanotubes), of the nano-structures. Togain a better understanding of the detrimental influence of the presenceof point defects atomic identification of occurring point defects is of vitalimportance. Up to the present the only known technique able to revealthe required atomic-scale information is electron spin resonance. Hence, inthis thesis we will use electron spin resonance to monitor, characterize, andhopefully identify occurring point defects in a range of materials of interestfor future developments in semiconductor nano-technology.

The outline of this thesis is as follows: Chapter 1 starts with a summaryof some elementary notions of the theory and practice of electron spinresonance experiments. Further a brief overview of the characteristics ofsome occurring defects is Si/SiO2 and in Si/high-κ structures is given.

In chapter 2 an extensive electron spin resonance study is presented offumed silica nanoparticles. Monitoring occurring point defects as a functionof post-formation heating and treatment revealed interesting atomic-scaleinformation concerning the particles’ network structure.

The third chapter reports on the observation of P-impurity related pointdefects in nm-thick P-implanted HfO2 films on (100)Si and in ZrO2 powder–two oxides prominent in current high-κ insulator research. It is shownthat the incorporation of P in these high-κ oxides results in ESR-activedefects possibly acting as hole traps. This finding is important in view ofthe observed enhanced dopant penetration through HfO2 layers during thenecessary dopant activation anneals.

The study of the nature and stability of the (100)Si/LaAlO3 interfaceis the subject of chapter 4. Here it is demonstrated that the interfaceis abrupt and stable for annealing up to about 800 ◦C. It is evidencedthat upon annealing in the range 800-860 ◦C a Si/SiO2-type interlayerstarts forming. Upon annealing at temperatures higher then 930 ◦C theinterlayer with SiOx nature is found to break up. The latter is possibly

Introduction 21

related crystallization and possibly silicate formation.In the last chapter the influence of ion implantation in amorphous bulk

SiO2 was studied. In cooperation with R. Weeks (Vanderbilt University,USA), R. Magruder (Belmont University, USA), and R. Weller (VanderbiltUniversity, USA) the densities of observed defects were compared to theoptical absorption bands around 4.8 and 5.3 eV . In this manner informationcould be obtained concerning the source of the optical absorptions.

This thesis ends with a summary and general conclusions of the experi-mental work performed.

22 Introduction

Chapter 1

Electron spin resonanceprobing of defects

1.1 Electron spin resonance: theory and experi-ment

When it comes to the atomic-scale characterization and identification ofdefects in solids, electron spin resonance (ESR) has proven to be a valuabletechnique. Of course only paramagnetic states can be studied with thistechnique, with a limited sensitivity –operated at low temperatures (4.2-100K) typically a detection limit of 1010 spins per Gauss line width is quoted.Continuous wave (cw) ESR experiments generally consist of measuring the(derivative) absorption of the microwave power (Pµ) at a fixed frequencyν as a function of an externally applied magnetic field induction ~B. Thetheory and practice of ESR has been accurately dealt with in numeroustextbooks [1, 2, 3] and review papers [4]. Here, we will restrict to a verybrief introduction to those ESR elements most commonly encountered insemiconductor ESR research.

1.1.1 Theory

The resonance condition

The energy of the interaction of a magnetic moment ~µ in a magnetic fieldinduction ~B may be written as

E = −~µ · ~B. (1.1)

23

24 Chapter 1: ESR probing of defects

The magnetic moments concerned in ESR originate from electrons, to whichthere are two contributions: the intrinsic angular moment ( ~µe) –the socalled spin– and the orbital angular moment ( ~µL). The magnetic momentarising from the electron spin can be written as

~µe = −g0µB ~S, (1.2)

where ~~S (~ is Planck’s constant h divided by 2π) is the electron spinangular momentum operator, µB ≡ |e|h

2me(e is the electron charge and me

the rest mass of the electron) the Bohr magneton, and g0 the so calledfree electron g factor. Experimentally as well as theoretically g0 has beendetermined to equal 2.0023193043718(75) [5]. The orbital angular momen-tum of an electron is formally written the same as Eq. 1.2, but without theproportionality factor g0,

~µL = −µB~L, (1.3)

where ~~L is the orbital angular momentum operator.For a spin system of total angular momentum ~J placed in an applied

magnetic field induction ~B the possible energies are quantized. The Zee-man levels can be written as

EMJ= gµBBMJ , (1.4)

with MJ = −J,−J + 1, ..., J − 1, J the magnetic quantum number. Theenergy levels are equidistant with an energy difference ∆E = gµBB betweentwo adjacent levels.

During an ESR experiment ’physical’ information about the spin sys-tem is obtained through inducing transitions between the Zeeman levels.According the basic theory of quantum mechanics, such transition can beaccomplished by a time dependent perturbation, e.g., an alternating mag-netic field ~B1 cos(2πνt), if hν = ∆E and ∆MJ = ±1. The resonancecondition then easily follows from Eq. 1.4 as

hν = gµBB. (1.5)

In Fig. 1.1 the Zeeman energies of Eq. 1.4 are diagrammed, for a single elec-tron with S=1/2, as a function of B together with the resonant absorptionof microwaves occurring when the resonance condition (Eq. 1.5) is fulfilled.

The effective spin Hamiltonian

Within the context of an ESR experiment, a spin system can be describedby a unique spin Hamiltonian (HS), i.e., a Hamiltonian build up of useful

1.1 ESR: theory and experiment 25

dPµ/d

B

Bmagnetic field

0 B rh g Bν µ=

0

12 BE g Bµ↓ =

0

12 BE g Bµ↑ = −

Ene

rgy

1/2

1/2 ∆B

∆Bpp

Br

Figure 1.1: Energy levels scheme as a function of magnetic field for the Zeemaninteraction of a single electron (S=1/2) together with the ESR resonance absorp-tion and its first derivative occurring at B = Br. Various indicated symbols, usedto describe signals, are discussed in the text.


terms only comprising spin operators and parameters involving quantitativeinformation about the nature of the defect under study. This is true when~L is quenched by the crystal field so that mixing of higher lying levels inthe ground state via spin-orbit coupling is small –a condition satisfied formost defects embedded in a solid, and certainly for all defects studied inthis thesis. Thus for the purpose of this work the full Hamiltonian can bereplaced by a reduced effective spin Hamiltonian

HS = Hez +HD +Hhf +Hnz, (1.6)

where Hez represents the operator of the electronic Zeeman interaction(0-1 cm−1), HD describes the spin-spin interactions (0-1 cm−1), Hhf thehyperfine (hf) interaction (0-102 cm−1), and Hnz represents the nuclearZeeman interaction (0-10−3 cm−1). We will now describe the theoreticalexpressions for the different terms in Eq. 1.6.

For a simple S=1/2, I=0 spin system the only non-zero interactionincluded in Eq. 1.6 is the electronic Zeeman interaction. The effective spinHamiltonian can be reduced to

HS = Hez = µB ~B · g · ~S. (1.7)

If the angular orbital momentum ~L is largely quenched, as assumed in theconstruction of the effective spin Hamiltonian, the remaining small residualmixing of the exited state |Ψn〉 spatial wave functions of the unpaired spinin the ground state |Ψ0〉 wave function by the spin-orbit coupling coefficientλ can be described by replacement of the real spin by an ”effective” sping · ~S/g0, effectuated by the replacement of the free electron g value g0 bya g matrix. Using perturbation theory, accurate to second order in thespin-orbit coupling, the g matrix becomes

g = g01 + 2λΛ, (1.8)

where 1 the unitary matrix. For unpaired spins localized on a single atom(or a cluster of like atoms), the elements of the symmetric matrix Λ aregiven by

Λij =∑n 6=0

〈Ψ0|Li|Ψn〉〈Ψn|Lj |Ψ0〉E0 − En

(1.9)

where E0 and En are the energy levels of the ground state and the excitedstate of the unpaired electron, respectively. From Eq. 1.9 it is evident thatthe symmetry of Ψ0 and Ψn is projected into the symmetry of the g matrix.This way, the g matrix can reveal information regarding the electronic


symmetry of the paramagnetic defect and provides a unique fingerprint ofthe center.

In general the g factor forms a symmetric matrix with six components.It is, however, always possible to find the principal axes (x, y, z) where theg matrix is diagonal

g =

gx 0 00 gy 00 0 gz

(1.10)

For an arbitrary orientation of a crystal in a magnetic field one obtains aresonance characterized by the g factor

g = (g2x cos2 θx + g2

y cos2 θy + g2z cos2 θz)1/2 (1.11)

where, for example θx is the angle between the x axis and the magnetic fielddirection and cos θx is called the direction cosine of x. The three directioncosines obey the relation

cos2 θx + cos2 θy + cos2 θz = 1 (1.12)

In spherical coordinates Eq. 1.11 assumes the form

g = (g2x sin2 θ cos2 φ+ g2

y sin2 θ sin2 φ+ g2z cos2 θ)1/2 (1.13)

It is frequently found that the g matrix has axial symmetry, in which case

g‖ = gz (1.14)g⊥ = gx = gz

where the z axis is taken as the symmetry axis. For this case Eq. 1.13becomes

g = (g2⊥ sin2 θ + g2

‖ cos2 θ)1/2 (1.15)

where θ is the angle between the symmetry axis (along g‖) and the magneticfield direction.

For S >1/2 spin systems the spin-spin interaction comes into play andmust be included in the effective spin Hamiltonian

HS = Hez +HD = µB ~B · g · ~S + ~S · D · ~S. (1.16)

The last term is often referred to as the fine structure or zero-field splittingand is most frequently expressed in its alternative form

HD = D[Sz2 − 13S(S + 1)] + E(Sx2 − Sy2), (1.17)


withD = Dzz −

12

(Dxx +Dyy),

andE =

12

(Dxx −Dyy).

For I 6=0 the nuclear Zeeman interaction and the hf interaction –the in-teraction of the unpaired electron with nearby nuclear magnetic moments–need to be included in the spin Hamiltonian describing the magnetic behav-ior of the ground state of the spin system. The effective spin Hamiltonianfor an S=1/2 electron interaction with a magnetic nucleus is

HS = Hez +Hhf +Hnz = µB ~B · g · ~S + ~S · A · ~I − gNµN ~B · ~I, (1.18)

where the symbols ~I, A, gN , and µN represent the nuclear spin vectoroperator, the hf tensor, the nuclear g factor, and the nuclear magneton.The last term in Eq. 1.18, representing the nuclear Zeeman interaction,can often be neglected except when the nucleus in question has a largemagnetic moment (e.g., the H atom). The middle term, representing thehf interaction, is replaced by a sum of similar terms when the electronspin interacts with several magnetic nuclei at different neighboring sites ordifferent ~I.

The hf interaction can be divided in two parts: the isotropic Fermi-contact interaction and the anisotropic electron nuclear dipole-dipole cou-pling. Consequently the hf tensor A can be written as

A = aiso1 + T , (1.19)

the sum of the isotropic hf interaction tensor aiso1 and the anisotropic hfinteraction tensor T . The Fermi-contact Hamiltonian Hf can be written as

Hf = aiso~S · ~I, (1.20)

whereaiso =

23µ0geµBgNµN |Ψ0(0)|2,

µ0 the permeability of free space, and Ψ0(0) the ground state electron wavefunction at the interacting nucleus. From Eq. 1.20 it is evident that themeasurement of the isotropic hf interaction may result in the determinationof the probability density of the electron at the site of the interactive nucleusif the chemical identity of the nucleus is known. Obviously, there is onlyan aiso contribution for s-state wave functions of the magnetic nucleus.


The electron nuclear dipole-dipole coupling Hamiltonian Hdd can bewritten as

Hdd =µ0

4πgeµBgNµN [

(3~S · ~r)(~r · ~I)r5

−~S · ~Ir3

] = ~S · T · ~I. (1.21)

Detection of occurring hf structures is, without doubt, the most powerfulESR observation. From the hf spectra the nuclear spin of the interactingnuclei and their (natural) abundance can be determined together with A.In this way it is possible to identify the atoms involved, their arrangement,and, in favorable cases, distances. In conjunction with theoretical mod-eling, the observation of the hf structures can often lead to full atomicidentification of the defect.

Linear Combination of Atomic Orbitals analysis of the hyperfineinteraction

The LCAO (Linear Combination of Atomic Orbitals) approximation isbased on the expansion of the molecular orbital Ψ0 containing the un-paired electron in terms of atomic wave functions that consist of a linearcombination of atomic orbitals φj . If all N atoms on which the groundstate wave function Ψ0 has an appreciable probability density are identicaland their atomic functions on each consist of M atomic orbitals then

Ψ0 =N∑i=1

ηi[M∑j=1

αjφj ], (1.22)

where ηi is the localization of the wave function on atom i and αj is its φjorbital character.

For the analysis of the hf interaction in this thesis the number of atomsis restricted to those causing a detectable hf splitting. The expansion isalso restricted to s and p orbitals only. This is a good approximation forthe analysis of defects in solids with top valence and conduction bands areconstructed from sp3 hybrid orbitals. Thus Ψ0 can be approximated as

Ψ0 =N∑i=1

ηi[αiφns + βiφnp], (1.23)

with the normalization conditions

∀i, i = 1...N : α2i + β2

i = 1

30 Chapter 1: ESR probing of defects∑i

η2i = 1.

A last assumption is that the wave function density on atom i totallyarises from the atomic wave function of atom i. Any contributions on atomi from the atomic wave function of an atom j 6= i are neglected. This lastassumption and the restriction to s and p orbitals imply that the hf matrixA is axially symmetric. In general, principal values of A can be writtenin terms of an isotropic (Aiso) part (s part: Fermi contact interaction), ananisotropic (b) part (p part: dipolar interaction), and an orthorhombic (c)part (deviation from perfect sp character):

A1 = Aiso + 2b (1.24)A2 = Aiso − b+ c

A3 = Aiso − b− c.

Within the assumption of perfect sp3 hybridization (c=0), the combinationof Eqs. 1.18, 1.20, 1.21, 1.23, and 1.24 renders:

| < φns | Ψ0 > |2 = η2α2 =AisoAs

, (1.25)

| < φnp | Ψ0 > |2 = η2β2 =b

Ap,

1 = α2 + β2,

where As and Ap are, respectively, the atomic s and p-state hf couplingconstants, inferred from theory. Thus the LCAO analysis of the hf inter-action presents a simple method to investigate the localization (η2) of theelectron wave function on an atom and the s (α2) and p (β2) character ofthe orbital composition of the unpaired electron.

As evident from Eq. 1.25, it should be noted that the inferred valuesfor α2, β2, and η2 will depend on the used atomic set for As and Ap.Some of the atomic hf coupling constants existing in literature are listed inTable 1.1 for Si and P. Comparison of the numbers in Table 1.1 reveals thelimited accuracy of the calculations (∼ 30%). It is therefore necessary, whencomparing various sources of hybridization and localization parameters, tomake sure whether the data were all obtained from interpretations basedon the same set of atomic hf coupling constants. In this thesis the valuescalculated by Morton and Preston [8] are used, as those are still regardedas the most accurate. Results from literature presented in this work are,if necessary, recalculated from the hf matrix using the atomic set fromRef. [8].


Table 1.1: List of various theoretical estimates of As and Ap, the hyperfine cou-pling constants for 3s and 3p electrons on atomic silicon and phosphorus.

Atom As (MHz) Ap (MHz) Ref.Si 4788 [6]

95 [7]4594 114 [8]3380 86 [9]4191 106 [10]

P 13306 367 [8]11175 308 [11]

1.1.2 Practice

Detection modes

In general first derivatives of absorption curves (dPµ/dB) are detected in cwESR, due to the use of lock-in detection techniques to improve sensitivity.The modulation amplitude Bm of the applied magnetic field B and incidentmicrowave power Pµ were restricted to levels not causing (noticeable) signaldistortion. The latter is very pertinent with respect to inferring reliableESR parameters of highly saturable point defects, such as most defectsencountered in this thesis. In the case of highly saturable defect centersit is sometimes just impossible to detect these undistorted as the properspectrometer settings cannot be attained due to the loss of sensitivity whendecreasing Pµ. A possible solution may then come from switching to thesecond harmonic phase-quadrature (out-of-phase) mode [12] using relativelyhigh Bm and Pµ, as demonstrated in chapters 2 and 3. Even though thesecond harmonic detection technique has not been finalized theoretically,empirical evidence has been provided that the observed spectra resemblethe direct absorption shapes [13, 14, 15]. The down side is that quantitativedetermination of defect densities from second harmonic saturation spectrais not straightforward and requires extreme care.

To enhance the signal-to-noise ratio, in both modes the spectrometersignal averaging capabilities are extensively used: if necessary typically upto several hundred field scans were added.


Physical parameters

Depending on sample and measurement conditions, various physical param-eters occurring in Eq. 1.6 can be obtained by interpretation of experimen-tally observed ESR spectra. The main parameters one attempts to measureare the three principal-axis components and directions of the A tensor andof the g matrix. For accurate g value calibration use was made, in thiswork, of a calibrated Si:P marker [16] sample [S=1/2, g(100 K)=1.99891]or a Li:F marker [17] sample [S=1/2, g=2.00229] co-mounted with the stud-ied sample and recorded in one trace. Chapter 3 of this work deals with thedetailed analysis of the hf interactions leading to the identification of theobserved defect centers. In chapter 2 the observation of changes in g andA, in conjunction with elementary theoretical modeling, reveals changes inthe network structure.

In general g and A are not isotropic and the observed g value and hfsplitting depend on the orientation of g and A, respectively, relative to thedirection of ~B. The orientation and symmetry of g and A reflect those ofthe involved electronic wave function, as can be seen from Eqs. 1.8, 1.20,and 1.21, which is co-determined by the defect’s local surrounding. Espe-cially in a crystalline environment, where defects occur in discrete orien-tations determined by the lattice symmetry, useful information concerningdefect and lattice orientation may be obtained. In disordered media, suchas poly-crystalline or amorphous materials, however, anisotropic spectralproperties are continuously averaged over all orientations of the defect,leading to the observation of less informative, smeared out spectra. If allorientations of the local coordinate system of the defect with respect to themagnetic field are present in the sample, the resulting ESR spectrum willthen be a so called powder spectrum –of importance in chapters 2, 3, and 5.The powder spectrum is a summation of all individual spectra over all pos-sible orientations of the principal axes of the defect relative to the directionof the magnetic field, each multiplied with its transition probability. Manyapproximations exist for the calculation of such a powder spectrum. Ex-amples of calculated powder spectra for an axial and orthorhombic systemare illustrated in Fig. 1.2.

Other salient spectral features are line width and line shape. The firstderivative absorption signal is characterized by the peak-to-peak width∆Bpp, the added result of various line broadening mechanisms. The mostfrequently encountered basic absorption shapes are the Gaussian, Lorentzian,and Voigt line, i.e. where the latter is the convolution of a Gaussian andLorentzian line. The full width at half the amplitude of the absorption


3

hνg β

2

hνg β

1

hνg β

dPµ/d

B

magnetic field

orthorombic system(a) (b)

2 3g =g =g⊥

1 //g =g

dPµ/d

B

magnetic field

axial system

Figure 1.2: Illustration of first derivative powder pattern spectra (Gaussian con-voluting line shape) for a S=1/2 system exhibiting (a) orthorhombic (the principalvalues of g are g1 6= g2 6= g3) and (b) axial symmetry (g1 6= g2 = g3, or g1 = g‖and g2 = g3 = g⊥, with g‖>g⊥).

line (∆B) is related to ∆Bpp through the expressions ∆B =√

2ln2∆Bppand ∆B =

√3∆Bpp for a Gaussian and Lorentzian line shape, respectively.

The line shape can be typified by the line shape factor

κls ≡2I

App∆B2pp

, (1.26)

where App is the peak-to-peak height of the signal, equal to twice the signalamplitude, and I the signal intensity, defined as the area under the absorp-tion curve. The line shape factors for Lorentzian, Voigt, and Gaussian lineshapes are κls=3.63, 1.03<κls<3.63, and κls=1.03, respectively.

The ESR parameters are often inferred from the observed spectra throughspectra simulations. The accuracy of the inferred g and hf values can beimproved through independent fitting of the spectra of ESR measurementsat different microwave frequencies. When this can be accomplished usingone set (within experimental error) of g and A, the inferred data can beregarded as highly reliable. It should be noted that in cases of ’large’ hfvalues (A &40 G) the fitting needs to be done using a code based on exactmatrix diagonalization incorporating the Breit-Rabi formula [18].

Defect density calibration

The signal intensity I is, apart from obvious experimental factors such as,e.g., the incident microwave power Pµ and the applied modulation fieldBm, proportional to the magnetic susceptibility χ. As a result I displaysa dependence on the number of spins in the sample. In this work defect


densities are determined by comparing IS for the sample under study withIM from a calibrated co-mounted marker sample. For this purpose thesame markers were used, i.e., Si:P and Li:F [16, 17], as for the g valuecalibration described before. A major difference between the two typesof markers is their paramagnetic behavior: Like most paramagnetic spinsystems, the χ of the Si:P marker displays a Curie-Weiss type behavior,i.e.,

χ ∼ 1T − TC

, (1.27)

where TC is the Curie temperature [TC(Si:P)=-2.7±0.3 K] [16]. For theLi:F marker the ESR signal originates from the conduction electrons of themetallic Li particles. Consequently, the marker exhibits a Pauli paramag-netic behavior giving rise to a χ essentially independent of the tempera-ture. All Li:F markers were calibrated at room temperature relatively toa Curie-Weiss paramagnetic primary marker (MgO:Cr3+), and then usedas intermediate spin marker with, however, always paying attention to itsspecific T-independent susceptibility.

When the ESR spectrum of an S=1/2 (spin) system exhibiting Curie-Weiss paramagnetism and g values close to the g value of the marker isrecorded –using the proper spectrometer settings– in the same trace as theco-mounted S=1/2 marker sample, the number of defects (spins) (NS) canbe obtained by

NS =IS

IMNM< B2

1 >M

< B21 >

S

T − TSCT + 2.7

, (1.28)

or

NS =IS

IMNM< B2

1 >M

< B21 >

S(T − TSC ), (1.29)

when comparing to the Si:P and Li:F marker, respectively. In Eqs. 1.28and 1.29 NM is the total number of spins in the co-mounted marker sampleand < B2

1 > represents the value of the square of the applied magneticmicrowave field component averaged over the sample. The assumptionmade in order to construct these equations are valid for all defects studiedin this thesis.

Correct signal intensities must be determined through double numericalintegration of the recorded first derivative ESR spectrum. However, thissometimes becomes inapplicable because of overlap and entanglement ofsignals originating from different types of defect centers. In that case re-liable intensities can only be obtained through reliable fitting of the ESR

1.2 Defects in the Si/SiO2 system 35

spectra and double numerical integration of the individual simulations ofthe different defect centers making up the measured ESR spectrum.

Spectrometers

The ESR measurements reported in this work were carried out employingthree experimental setups: One home built system operated at K-band(ν ∼ 20.5 GHz) frequencies, a commercial Q-band (ν ∼ 34 GHz) setup(Bruker EMX) and a commercial X-band (ν ∼ 9.2 GHz) setup (Jeol JES-FA 100). The systems show some similarities: all are equipped with acylindrical TE011 microwave reflection cavity, routinely driven in absorp-tion mode under conditions of slow adiabatic passage. Modulation of theexternally applied magnetic field induction ~B typically at 100 kHz, allowsphase sensitive signal detection.

The K-band spectrometer is equipped with a liquid He bath cryostatwhich allows variations of the measurement temperature T in the range1-300 K. Usually, however, it is operated at 4.2 K, i.e., the temperatureof liquid He at atmosphere pressure. The K-band cavity has a very highquality factor Q compared to the commercial setups: the typically loadedQ at ∼ 10 K was ∼ 10000. Moreover, the distinct differences in microwavecircuitry and related detection scheme, enabled the K-band spectrometerto be operated, still with superb sensitivity, at much lower Pµ –below nWrange– than the Q or X-band setups. This quality is crucial for the success-ful conventional absorption-derivative detection of highly saturable defectcenters. The Q-band spectrometer is equipped with a He flow cryostatand operates generally at temperatures T > 20 K to room temperature.For the Q-band setup a maximum loaded Q of ∼ 4000 could be attained.The Q-band spectrometer can be operated at Pµ in the range 160 nW -160 mW . The Q-band setup was found to be the most suited for secondharmonic out-of-phase mode measurements. The X-band spectrometer isequipped with a N2 flow cryostat and operates generally at temperaturesT ∼ 120 - 470 K. A loaded Q of maximum ∼ 6000 could be attained andthe spectrometer can be operated at Pµ in the range 200 mW - 0.1 µW .

1.2 Defects in the Si/SiO2 system

For decades, the Si/SiO2 structure has dominated the semiconductor in-dustry. Some of the reasons are that this native insulator to Si providesthermodynamical and electrical stability, a large band gap ( ∼ 9 eV , mak-ing it an excellent electrical isolator), and an outstanding Si/SiO2 interface


(a)

(b)

Figure 1.3: Illustration of Moore’s law stating that the number of transistors ona chip doubles every 18 months: (a) Gordon Moore’s original graph from 1965,and (b) growth of transistor counts for Intel processors following Moore’s law upto the present.


because of the passivation behavior (very efficient possibility for passivationby H) of the inherently present interface defects. Up to now there is nointerface exhibiting electrical properties approaching those of the Si/SiO2

interface.Over the years improvement of the performance of integrated circuits

(ICs) was successfully achieved by scaling down the metal-oxide-semicon-ductor (MOS) entity, as illustrated in Fig. 1.3 by Moore’s law. Lateralscaling of the MOS devices and thus the gate oxide mandates compen-sation for the loss in areal gate capacitance. This may be realized byreducing the gate oxide thickness. For the SiO2 or the nitrided silicate(SiOxNy) gate dielectric down-scaling could be successfully accomplishedfor gate thicknesses down to the 1-nm range [19, 20]. There is, however,a fundamental limit to this scaling being the loss of bulk SiO2 propertiesbelow ∼7 A. From a technological point of view the ultimate failure ofstandard SiO2 as gate insulator is mainly due to issues such as intoler-able (tunneling) leakage currents, problems with intrinsic reliability, anddopant penetration. Consequently, the ongoing scaling of MOS based de-vices will demand, amongst others, the introduction of new wide band gapinsulators of dielectric constant κ higher than that of conventional SiO2 orSiOxNy (κSiO2 = 3.9, κSiOxNy = 3.9 − 7.5) in the near future, known asthe high-κ issue [21, 19]. In that manner a physically thicker oxide layer ofequivalent SiO2 electrical oxide thickness (EOT, defined as dEOT=dhigh−κκSiO2/κhigh−κ) can be used retaining the same gate capacitance but possi-bly reducing the leakage current and improving the gate reliability.

To set the stage for the discussion of Si/high-κ structures studied inChapter 3 and 4, an overview is given of the defects, relevant to this work,observed in Si/SiO2 structures. Even though it seems that the conventionalSi/SiO2-age will soon come to an end, the Si/SiO2 entity will remain of in-terest as it appears inevitably present in the Si/high-κ structures. On theother hand, the oxide itself has regained attention in the form of nm-sizedparticles for their potential applications in optical and electronic nanode-vices [22, 23]. (Chapter 2)

1.2.1 Pb-type interface defects

Three Si dangling-bond type Si/SiO2 interface defects have so far been sub-stantiated by ESR, commonly denoted as the Pb-type defects. It has beendemonstrated in conjunction with electrical measurements [25, 26] thatthese Pb-type defects are the dominant class of interface traps invariablyintroduced at the Si/SiO2 interface as a result of lattice mismatch between


O

Si

Pb

[111]

Figure 1.4: A schematic representation of the (111)Si/SiO2 interface with anincorporated Pb defect structure [24].

c-Si and a-SiO2 [27, 28, 29]. The different types of Si dangling-bond typedefects observed correlate with interface orientation, in registry with thecrystallinity of the underlying Si substrate.

At the (111)Si/SiO2 interface, the only type observed –specifically termedPb– was identified as trivalent interfacial Si (Si3 ≡ Si•, where the dot repre-sents an unpaired electron in a dangling Si sp3 hybrid) back bonded to threeSi atoms in the bulk. A schematic representation of the atomic structure ofthe (111) surface and the incorporated Pb defect are shown in Fig. 1.4. ThePb center was observed for the first time by Nishi [30]. Later, more extensivemeasurements performed by the group of Poindexter and Caplan [31, 32]linked the Pb densities to the densities of interface traps in Si/SiO2 capaci-tors and provided an accurate g map. Figure 1.5 illustrates the calculatedg map of the Pb center. The principal axis g values used are g‖=2.0013and g⊥=2.0086. The center exhibits C3v symmetry (axial along the [111]),for which four equivalent orientations in the Si lattice exist. Yet, only oneorientation along the [111] interface normal is generally observed [29]. Theother orientations, however, have been observed as well [33]. Conclusiveevidence for the atomic model of the Pb center was provided by the obser-vation of a strong 29Si (natural abundance 4.7%; I = 1/2) hf interaction


0 10 20 30 40 50 60 70 80 902.001

2.002

2.003

2.004

2.005

2.006

2.007

2.008

2.009

2.010

Pb

g va

lue

magnet angle Φ (deg)

[111] [211] [111] [ 211]

Figure 1.5: Angular g map of the Pb center observed at the (111)Si/SiO2 interfacefor ~B rotating in the (011) plane with respect to the interface normal ~n. The curverepresents the theoretical calculation for trigonal symmetry using the principal axisg values g‖=2.0013 and g⊥=2.0086. Only the experimentally observed branch isshown.

with a single Si site [24]. LCAO analysis of the hf tensor indicated that80% of the unpaired electron density is located on the central Si atom. Theorbital lobe pointing into the oxide was found to exhibit 12% s and 88% pcharacter, confirming its expected sp3 hybrid like character.

The technologically favored (100)Si/SiO2 interface exhibits two types ofdangling-bond type centers, termed Pb0 and Pb1, again first observed by thegroup of Poindexter and Caplan [32]. Their atomic assignment, however,remained long obscure mainly because of the lack of convincing ESR datadue to the inherently smaller defect densities (∼1/5 as compared to theoccurrence of Pb on (111) interfaces) and the spectral overlap. Later, mea-surements on thermal (100)Si/SiO2 predominantly exhibiting either the Pb0or Pb1 defect allowed accurate g mapping of both interface defects [34, 35].The calculated g map is shown in Fig. 1.6. The principal g values usedare g‖=2.00185 and g⊥=2.0081 for the Pb0 defect center and g1=2.0058,g2=2.00735, and g3=2.0022 for the Pb1 defect center. The experimentalevidence is that Pb0 is chemically identical to Pb, but now residing at mi-croscopically (111)-oriented Si/SiO2 facets [34, 36, 37]. In agreement, bothwere conclusively established as major systems of detrimental electrical in-terface traps [25, 26]. The Pb1 center is assigned to a distorted defected


0 10 20 30 40 50 60 70 80 902.001

2.002

2.003

2.004

2.005

2.006

2.007

2.008

2.009

2.010

2

1

1

12

g va

lue


Pb0

Pb1

1

[111] [211] [100] [011]

Figure 1.6: Angular g map of the Pb0 (black curves) and Pb1 (red curves) centersobserved at the (100)Si/SiO2 interface for ~B rotating in the (011) plane with re-spect to the interface normal ~n. The curves represents the theoretical calculationsfor trigonal symmetry using the principal axis g values g‖=2.00185 and g⊥=2.0081for the Pb0 defect center and for triclinic symmetry using the principal axis g val-ues g1=2.0058, g2=2.00735, and g3=2.0022 for the Pb1 defect center. Only theexperimentally observed branches are shown. The added numbers indicate relativebranch intensities.

interfacial Si-Si dimer (a ≡ Si—Si•=Si2 defect, where the long hyphen sym-bolizes a strained bond, with an approximately <211> oriented unpairedSi sp3 hybrid) [38]. An (100)Si/SiO2 interface with incorporated Pb0 andPb1 defects is illustrated in Fig. 1.7.

At the (110)Si/SiO2 interface also, only one type, the Pb variant, isobserved [27].

Thus all three variants were shown to be interfacial trivalent Si cen-ters [28, 38, 34], naturally occurring, for standard oxidation temperaturesTox (800-960 ◦C), in areal densities of [Pb] ∼ 5 × 1012 cm−2 [39, 40] and[Pb0], [Pb1] ∼ 1 × 1012 cm−2 [39]. Both Pb and Pb0 were demonstrated tobe adverse electrical interface traps [25, 26].

As a key characteristic, the thermochemical properties of the Pb-typedefects appear dominated by reversible interaction with hydrogen. Thedefects may be readily electrically inactivated (passivated) by heating (Tan& 230 ◦C) in molecular H2 pictured as chemical saturation of the unpairedsp3 Si hybrid by hydrogen [28, 41, 42, 43], denoted as SiH formation. Thechemical reaction is modeled as Pb + H2 → PbH + H↑, proceeding with


Instituut voor Nanoschaal Fysica en Chemie

O

Si

Pb0

Pb1

[100] [111]

Figure 1.7: A schematic representation of the (100)Si/SiO2 interface with incor-porated Pb0 and Pb1 defect structures [37, 38].

an activation energy Ef=1.51±0.04 eV [41, 42, 43]. Heating to Tan &500 ◦C results in dissociation of H, according to the reaction PbH → Pb+ H↑ turning the defects again into the ESR-active paramagnetic state[44, 41, 42, 43]. The latter reaction proceeds with an activation energy ofEd=2.83±0.2 eV [45].

1.2.2 Intrinsic defects in a-SiO2

A range of intrinsic defects have been reported in a-SiO2 including the E′-type defects, EX, the non-bridging oxygen hole center (NBOHC), andthe peroxy-radical (POR). In the following we will discuss those intrinsicdefects in a-SiO2 of relevance to this work. Models of the various defectsare discussed and compiled in Figs. 1.8, 1.9, 1.10, and 1.11.


Table 1.2: Comparison of experimental principal axis g matrix values of genericE′ centers.

E′1 [a] E′γ [b] E′α [b] E′β [b] E′δ [b] E′s [c]g1 2.0018 2.0018 2.0018 2.0018 2.0018 2.0018g2 2.0005 2.0006 2.0013 2.0004 2.0021 2.0003g3 2.0003 2.0003 1.9998 2.0004 2.0021 2.0003

[a] Observed in fast neutron irradiated crystalline quartz, see Ref. [47].[b] Observed in synthetic fused silica subjected to 100 keV x-rays at 77 K, seeRef. [48].[c] Observed in natural and synthetic quartz crushed in UHV at 300 K, seeRef. [49].

The E′γ center

The E′γ center is a subclass of the most widely studied radiation-induceddefect centers in a-SiO2, i.e., the so called E′-type defects. Most of theE′ defects share the same basic structural feature –the so called genericE′ entity– pictured as an O3 ≡Si• moiety, where • and ≡ represent anunpaired electron and three Si-O bonds respectively, with the unpairedelectron highly localized in an sp3-type hybrid orbital on the defect Si,back bonded to three O atoms. Their presence is closely related to the oxidequality. Beside the E′γ center, previous ESR studies have demonstrated theexistence of several variants of E′-type centers in a-SiO2 [4], that exhibitsubtle differences in their ESR signals as shown in Table 1.2. The modelsfor the different E′-type defects in a-SiO2 are still under discussion. Anoverview is presented in Ref. [46].

For the E′γ center Griscom et al. pointed out that the model is mostprobably essentially identical with that of the ’crystalline’ equivalent ofthis defect [14] –termed E′1 center– first observed in neutron irradiated α-quartz [50], because the ESR characteristics of these two types of E′ centersare very similar. The observed larger spread both in the nearly equal g2

and g3 components and in the isotropic 29Si hf coupling constant [51] ofthe E′γ center compared to the E′1 center were attributed to small randomvariations in the defect bond angles in the glass.

A detailed study of the E′1 center [47] revealed the principal values ofthe g matrix (see Table 1.2), and reported the observation of one strong hfinteraction (∼ 420 G doublet), which allowed identification of the unpaired


O

(a)

Si

(b)

+

+

Figure 1.8: A schematic representation of the suggested models of the E′γ center:(a) The conventional model [51, 53] and (b) the new BHODC model [54].

electron state as occupying a sp3 hybrid orbital on a single Si site. Inaddition, two more weak hf interactions were observed, also attributed to(more distant) Si atoms. The formation mechanism of the E′1 center inα-quartz, proposed by Feigl, Fowler, and Yip [52], is well accepted; thatis, a positive hole h+ trapped at a neutral oxygen monovacancy can belocalized on one silicon atom and then lower its energy by relaxing backinto the plane of its three remaining oxygen neighbors. This reaction canbe written as follows:

≡ Si− Si ≡ +h+ →≡ Si • (E′1)+ ≡ Si+, (1.30)

where • and ≡ represent an unpaired electron and three Si-O bonds, respec-tively. Importantly, as modeled, the E′1 center in its paramagnetic statewould thus be a positively charged entity. This model was subsequentlyrefined by Rudra and Fowler [53], who suggested that the relaxation of thepositively charged silicon is further stabilized by forming an additional weakbackward Si—O bond (”backward puckering” through the plane formedby the three back bonded O atoms) with a nearby oxygen atom, makingthe oxygen atom threefold coordinated. The latter model is illustrated inFig. 1.8 (a).

More recently, Uchino et al. questioned the assumption that the E′γcenter in a-SiO2 has the same microscopic origin as the E′1 center in α-quartz [54, 55]. Based on density functional theory (DFT) calculations, adifferent model –named the bridged hole-trapping oxygen deficiency center(BHODC)– was proposed, as shown in Fig. 1.8 (b). In this model a para-magnetic silicon shares a common oxygen with a neighboring hole-trappingsilicon. It is, however, important to notice that this model concerns one ofthe more theoretical suggestions, that still need experimental verification.


Both models described here hold that the E′ center is positively charged.Over the years various experimental studies have questioned this assumedpositive charge state of the paramagnetic E′ center in thermal SiO2 onSi [56, 57, 58]. Based on the comparison between the densities of positivecharge and paramagnetic E′ centers generated in thermal SiO2 on Si byVUV irradiation at different electrical field strengths in the oxide, Afanas’evand Stesmans [59] have demonstrated that an uncharged E′γ center doesexist in amorphous thermal SiO2 on Si and proposed that the E′γ centermainly involved the O≡Si• moiety only, i.e., the positively charged coun-terpart is absent. The E′γ system here operates as a potential H-storagesystem (O≡Si–H formation).

The S center

Holzenkampfer et al. carried out an ESR study on thin amorphous SiOx

layers deposited by electron-beam evaporation of Si under different oxygenpressures, where x was varied between 0 and 2 [60]. After irradiation withHe+ ions, a broad resonance line was observed which the authors inter-preted as a superposition of three singlet lines having different g values –inthe range ∼2.0060 to ∼2.0008– and line widths. From the variation of theeffective g value with the oxygen stoichiometry x, they inferred that thehigher g values are associated with regions of the network which are pro-gressively more oxygen deficient. In the samples with x ≥ 0.8 a narrow res-onance line centered near g∼2.001 was observed in increasing intensity withrising oxygen content. Based on the signal’s observed g matrix anisotropy,the signal was identified as the E′γ center. The broad resonances were in-terpreted in terms of superpositions of three additional signals arising fromE′ or Pb variants wherein the defected silicon is bonded to one oxygen andtwo silicons, two oxygens and one silicon, or three silicons (the latter beingthe Pb defect.

Si

O

Figure 1.9: A schematic representation of the suggested model of the S center.


Later, Griscom et al. observed a singlet (g∼2.0030; ∆Bpp∼2.6 G) inγ-irradiated P2O5-SiO2 glasses annealed at Tan≥700 K, the correspondingdefect of which they labeled the S center [61]. Within the framework of thework of Holzenkampfer [60], they suggested the S center to correspond tothe centers ”intermediate” between the E′ and Pb defects, namely defectsof the type OSi2≡Si• and/or O2Si≡Si•, as illustrated in Fig. 1.9.

Stesmans and Afanas’ev for the first time observed a signal similar tothat of the S center –characteristic isotropic signal at g(4.2 K)=2.0028 and∆Bpp=4.5-5.8 G– in thermal SiO2 on (100)Si subjected to post-oxidationvacuum annealing at Tan≥950 ◦C [62]. The S center generation dur-ing degradation was found attendant with the appearance of other oxidepoint defects: EX, E′γ , and E′δ [62, 63, 64]. A similar center (g=2.0028;∆Bpp=3.6-6 G) was also observed in fused silica upon the interaction withgaseous SiO during annealing at ∼1140 ◦C [65]. Later two hf doublets(162±5 G and 279±2 G) could be discerned in thermal SiO2 on Si subjectedto 1 h vacuum annealing at ∼1250 ◦C and additional VUV irradiation [64].These doublets were centered at the (central) Zeeman signal of the S center,and accordingly, interpreted as corresponding hf structure. The spectral in-tensity of the outer doublet appeared somewhat larger as compared to theinner doublet; However, given the large experimental uncertainty, the in-tensity of these might well be equal within the experimental error. Furtherinterpretation required more discriminative experimental information.

The experimental breakthrough realized by observation of two addi-tional doublets, in the S defect ESR spectrum, assigned as hf structure,has stimulated further theoretical work. Based on generalized-gradientDFT work [66], the defect has been identified with the Si≡SiO2 center,where the two doublets could be successfully modeled: The principal outerdoublet would arise from the defect Si atom, while the weaker inner doubletwould arise from the delocalized spin density on the Si atom in the firstneighbor shell of the defect Si atom.

The EX center

Another defect of interest for the current work is EX: An SiO2-associateddefect generally observed in Si/SiO2-entities after thermal treatment. TheEX center was first reported [67] for as-prepared SiO2 thermally grown on(111)Si in dry O2 at 700-850 ◦C. Using K-band ESR at low temperatures anarrow isotropic signal was observed with a g value of g=2.00246±0.00003and line width ∆Bpp=1.0±0.05G together with a 29Si hf doublet of splittingA=16.1±0.1 G. Later Stesmans et al. [69] reported the observation of EX


O

Si

Figure 1.10: A schematic representation of the proposed model of the EX cen-ter [68].

in dry thermal SiO2 in (100) and (111)Si/SiO2 entities. From this study itwas shown that the EX areal defect density depends on the grown oxidethickness (dox), that EX is only detected from dox ∼ 70 A onwards witha maximum density at dox ∼ 125 A and is mainly located in the top 45A of the oxide layer. The detection of part of the 17O hf structures ofEX in 17O enriched (111)Si/SiO2 structures [68] resulted in the suggestionof a preliminary model where EX consists of a hole delocalized over fouroxygen dangling bonds formally at the site of a Si vacancy. In anotherview, it has been described as an agglomorate of four oxygen-related holecenters (OHC’s) [70]. A schematic representation of the suggested modelfor the EX center is shown in Fig. 1.10.

Over the years EX has been observed in a variety of Si/SiO2 entitiesand SiO2 samples such as, e.g., oxidized porous Si layers [71, 70, 36], ther-mally degraded Si/SiO2 [63], silicon nanowires [72, 73, 74], ultra fine siliconparticles [75], and in stacks of nm-thick layers of SiOx, Al2O3, ZrO2, andHfO2 on (100)Si [76, 77, 78]. Dohi et al. suggested the presence of twodifferent variants in ultra fine Si particles [75], exhibiting the same g valuebut a different saturation behavior, which they labeled EXL and EXH .This work further suggested that the EXH center observed upon anneal-ing of the particles in vacuum at 900-1000 ◦C is similar to the EX centerdescribed above, and that the EXL center observed upon annealing at 600-700 ◦C is identical to the defect reported by Kusumoto [79] in Si powderscrushed in room ambient which Dohi et al. suggest to be related to the


SiO

(a) (b)

Figure 1.11: A schematic representation of the proposed generic entity of twoOHCs: (a) the NBOHC [80], and (b) the POR [81].

Si-OH structure; however, no detailed atomic model has been provided.

The oxygen associated hole centers

For reasons of completeness we will briefly discuss two types of OHCs: thePOR which is relatively more prevalent in silicas with a low OH content andthe NBOHC which tends to dominate in silicas with a high OH content [4].The latter is attributed to ≡Si-O• species, as illustrated in Fig. 1.11 (a).Originally it was isolated by ESR in γ-irradiated fused silica with a highOH content [80] and a first confirmation of this model was obtained byStapelbroek et al. [82] who measured the unpaired electron hf interactionwith 17O nuclei in an isotopically enriched sample. Further experimentalevidence was provided by the observation of the hf interaction with justone silicon in a 29Si enriched sample [83].

The POR was identified as a Si-O-O• entity in a study on 36 % 17Oenriched neutron irradiated silica glass [81]. The suggested model is pre-sented in Fig. 1.11 (b). This assignment was later corroborated by a studyof the same defect in 29Si enriched samples, which, in agreement with themodel, demonstrated the hf interaction with a single silicon atom [83].

1.2.3 Extrinsic defects in a-SiO2

In addition to the intrinsic point defects various extrinsic, or impurity re-lated, point defects have been observed in a-SiO2 [4]. In undoped irradiateda-SiO2 atomic hydrogen H0 is one of the most commonly encountered impu-rity associated defects, although it should be specified as a low temperature


H

O

Si

Figure 1.12: A schematic representation of the generic entity of the 73-G doublet.

defect. As generally created by ionizing radiation, it remains relatively sta-ble below ∼100 K, but it anneals out rapidly above ∼130 K, mainly as aresult of dimerization (H2 formation). Its ESR resonance signal is made upof a pair of sharp lines centered on g=2.00 with a splitting of ∼505 G dueto the hf interaction with 1H (99.98% natural abundance, I=1/2).

Another H-impurity related defect observed in various types of undopeda-SiO2 [15, 84] is the so-called 73-G doublet attributed to an E′ type defectwith a H atom in the back bond instead of O, as shown in Fig. 1.12.

In doped a-SiO2 and in silicate glasses, the number of impurity relateddefects observed by ESR is almost unlimited. Here we will focus on P-associated point defects in SiO2 as their discussion will constitute a basis forthe analysis of P-associated defects in high-κ oxides, presented in chapter 3.

P-associated defects in SiO2-based materials

Uchida et al. succeeded in the incorporation of P ions in hydrothermallygrown quartz crystals [85, 86]. Upon x or γ-irradiation of the doped crys-tals, ESR measurements performed at ∼40, 120, and 290 K revealed threesets of 31P hf lines labeled P (I), P (II), and P (A), where P (I) and P (II)were observed as separate ESR signals at low temperatures (T < 140 K)and P (A) appears at high T . It was demonstrated that the three sets ofresonances are separate but related appearances of an unpaired electronlocalized at a pentavalent P ion which occupies substitutionally a Si site:P (I) is the ground state of [PO4]0, P (II) is the first accessible excited stateof P (I), and P (A) is the dynamic average. Based on the symmetry proper-ties a structural model was proposed as presented schematically in Fig. 1.13(a), introducing an appreciable displacement of the P nucleus. Calculationsbased on DFT of the hf parameters agree well with the experimental val-ues and also indicated that the incorporation of P in the c-SiO2 networkresults in a significant perturbation of the α-quartz structure surrounding


Table 1.3: Overview of the ESR data in the literature on the different P-associateddefect centers observed in SiO2-based glasses and P-doped c-SiO2 together with thecorresponding principal values of the g and A matrices. Results are obtained atX-band frequency.

Defect label Oxide and damage g A(G)P1 10P2O5-90SiO2 g1=2.002 A1=1030

x- or γ-irradiated [a] g2=1.999 A2=850g3=1.999 A3=850

PSG films on Si g1=2.003 A1=989VUV irradiated [b] g2=1.998 A2=789

g3=1.998 A3=785P2 10P2O5-90SiO2 g1=2.001 A1=1300

x- or γ-irradiated [a] g2=2.001 A2=1150g3=2.001 A3=1150

PSG films on Si g1=2.010 A1=1360VUV irradiated [b] g2=1.980 A2=1120

g3=1.980 A3=1120P (I)-variant of P-doped α-quartz g1=2.0012 A1=1228.93

P2 [d] x- or γ-irradiated [c] g2=2.0032 A2=1086.86g3=1.9991 A3=1074.84

P (II)-variant of g1=2.0013 A1=1159.72P2 [d] g2=2.0034 A2=1025.14

g3=1.9991 A3=1012.22P (A)-variant of g1=2.0010 A1=1139.02

P2 [d] g2=2.0025 A2=1120.99g3=2.0003 A3=1057.79

P4 10P2O5-90SiO2 g1=2.0014 A1=355.2x- or γ-irradiated [a] g2=1.9989 A2=-43.8

g3=1.9989 A3=-43.8PSG films on Si g1=2.016 A1=274.5

VUV irradiated [b] g2=1.9989 A2=-43.5g3=1.9989 A3=-43.5

POHCs 10P2O5-90SiO2 g1=2.0179 A1=54x- or γ-irradiated [a] g2=2.0097 A2=52

g3=2.0075 A3=48PSG films on Si g1=2.0185 A1=54x-irradiated [e] g2=2.0115 A2=49

g3=2.0082 A3=47POHCm PSG films on Si g1=2.0514 A1=55

x-irradiated [e] g2=2.0079 A2=50g3=2.0032 A3=44

[a] See Refs. [4, 61].[b] Deposited by pressure induced CVD, see Ref. [87].[c] See Ref. [85, 86].[d] P (I) is ascribed to the ground state of P2, P (II) is the first accessible excited state of P (I),and P (A) is the dynamic average. P (I) and P (II) are observed as separate ESR signals at lowtemperatures (T < 140 K) and P (A) appears at high T [85, 86].[e] Deposited by subatmospheric CVD, see Ref. [88].


Figure 1.13: Overview of the conceptual models proposed in literature for theformation of P-associated defects in SiO2-based glasses: (a) P (I), P2 [85, 86], (b)P1 [61], (c) P4 [61], (d) POHCs [61], and (e) POHCm [61].


the impurity [89, 90].Griscom et al. reported on the observation of four different 31P dou-

blets in glassy 10P2O5-90SiO2 subjected to x- or γ-irradiation which theylabeled P1, P2, P4, and POHC [61]. The nomenclature P1 and P2 wastaken from Weeks and Bray [91] who studied P-associated defects in γ-irradiated P2O5 and alkali phosphate glasses as these defects appeared tobe essentially identical to those observed in P-doped SiO2. The P1 andP2 centers are characterized by a very large hf splitting (Aiso∼910 G and∼1200 G, respectively; cf. Table 1.3 for Aiso=A1+A2+A3

3 ). The P1 defectis suggested to be the phosphorous analogue of the Si E′ center in SiO2

presumably formed by trapping a hole on a three fold coordinated P atomsubstitutionally occupying a three fold coordinated Si atom at an E′ centersite. The proposed steric model is shown in Fig. 1.13 (b). The P2 cen-ter appeared to be identical to the [PO4]0 center reported by Uchida etal. [85, 86].

The P4 and POHC defects both exhibit a much smaller hf splitting thanP1 and P2 (cf. Table 1.3). The P4 center has been suggested to originatefrom the three fold coordinated P precursor site as P1 but is formed bytrapping an electron as depicted in Fig. 1.13 (c). Two different variantsof the POHC (phosphorus oxygen hole center) were observed: A stablevariant (POHCs) and a metastable variant (POHCm) only observed atlow temperatures (T ≤ 300 K). The POHCs was fully characterized andsuggested to be a hole trapped on a pair of non bridging oxygens bondedto the same phosphorous, see Fig. 1.13 (d). The POHCm was tentativelyassigned to a hole metastably trapped on the lone non-bridging oxygen, asdepicted in Fig. 1.13 (e).

The P-associated defect centers have not only been observed in bulkP-doped silica glass but also in thin films of SiO2-based glasses on Si [92,13, 87, 88]. Warren et al. observed the POHCs in phosphosilicate (PSG)and borophosphosilicate (BPSG) dielectrics deposited on Si by chemicalvapor deposition (CVD) subjected to VUV, x-ray irradiation, or hole injec-tion [92]. Based on the charge trapping behavior of the defect, a differentmodel was proposed for the POHCs where the spin active state would bepositively charged instead of neutral. This alternative model is the sameas the model proposed by Griscom et al. for the metastable variant ofthe POHC. Later on Fanciulli et al. rejected this alternate model for thePOHCs and retained the model proposed by Griscom et al. based onthe combination of DFT calculations and UV-Raman and ESR measure-ments on PSG films deposited on Si in a single chamber using subatmo-


TheoryElectronic structure of defects in a-SiO2

0

5

10

Ev

Ene

rgy

(eV

) Ec

≡Si

•

•

• •

≡P

P-impurities introduce defect states in band gap

Robertson Phys. Tech. a-SiO2, Les Arcs (1987)

P2

Figure 1.14: Calculated energy levels, based on DFT, associated with variousdefect configurations located in the band gap (∼9 eV) of SiO2. Defects showninclude the silicon dangling bond (≡Si), the P2 center, and the precusor of the P1

and P4 defects (≡P).

spheric CVD, and subsequently subjected to x-ray irradiation [88]. Boththe POHCs and POHCm centers were observed and modeled. In a differ-ent work, in addition to the POHCs, the P1 and P2 centers were observedin atmospheric pressure CVD deposited BPSG films on Si subjected toVUV irradiation, electron, and/or hole injection [13]. The P1, P2, and P4

defect centers have been observed as well in PSG films deposited on Si bypressure induced CVD, after subjection to VUV irradiation [87].

Theoretical work based on DFT studies showed that the introductionof P impurities in silica introduces defect states in the band gap, as shownin Fig. 1.14. In particular, the P2 defect has been assigned to be a holetrap [93].

1.3 Defects in the Si/high-κ system

As outlined in the International Technology Roadmap for semiconductors(ITRS) [21], continued scaling of complementary metal-oxide-semiconductor(MOS) technology will require the replacement of the conventional SiO2

gate insulator by an alternative dielectric of higher κ in the near termyears. The projected advancement in technology nodes and the relatedscaling of some of the device dimensions over the predicted years of pro-duction are shown in Table 1.4. As already outlined, high-κ gate dielectricmetal oxides permit the usage of physically thicker gate dielectrics to obtainthe same effective capacitance as with thinner SiO2 gate dielectrics, while

1.3 Defects in the Si/high-κ system 53

Table 1.4: Technology nodes, gate lengths, and gate dielectric EOTs for futurehigh performance applications [21] in current MOSFET-based technology.

Year of 2007 2010 2013 2016productionTechnology 65 45 32 22node (nm)

Physical gate 25 18 13 9length (nm)EOT (nm) 0.8-0.9 0.7 0.6 0.5

significantly reducing the probability for direct tunneling [19]. For thatpurpose, a large variety of (metal) oxides have been investigated as possi-ble candidates, including metal oxides in their simplest form such as HfO2,ZrO2, Al2O3, La2O3, and others [19, 94], as well as more complex varietiessuch as silicates, aluminates, pseudo binary mixtures and multi metal oxidecompounds. Currently, HfO2 or better stated HfO2-based composites (e.g.,HfSixOyNz), have emerged as most promising future gate oxides [21, 95].

In order to be successful the replacing oxide must meet various require-ments [21, 19, 20]: The dielectric constant should be somewhere between10 and 30, the high-κ oxide must be an insulator with a band gap over 5 eVand exhibiting sufficient band offsets with Si, it should have a high qualityinterface with Si and a low interface trap defect density, it should exhibitgood thermodynamic stability with Si, and good dopant barrier properties.

The various candidate materials have been intensively studied by nu-merous material characterization techniques providing a wealth of usefulinformation on the microstructure, bonding chemistry, and composition ofthe newly composed Si/high-κ stacks [96, 97, 98, 99, 100, 101, 102, 103, 104,105, 106, 107]. Also, using electrical techniques such as capacitance-voltage(C-V) and current-voltage (I-V), the investigation of the electrical perfor-mance (transport, charge trapping, and presence of detrimentally activeinterface traps) has received much attention [108, 109, 110, 111]. However,when it comes to assessing the atomic nature of these electrically activetraps these techniques, superb as they are, inherently fall short. That in-formation can in principle be obtained by ESR probing of occurring pointdefects at the Si/high-κ interface and in the high-κ oxide.


1.3.1 The Si/high-κ interface

The first ESR study of paramagnetic point defects at the interface of aSi/high-κ structure was conducted by Stesmans and Afanas’ev [112, 113].The initial K-band work studied (100)Si/SiOx/ZrO2 and (100)Si/Al2O3/ZrO2

stacks (SiOx refers to nonstoiciometric Si dioxide, with 1≤x≤2) with nm-thin layers grown at 300 ◦C by atomic layer CVD (ALCVD) using trimethy-laluminium, ZrCl4, and H2O as precursors. The only defects observed in allstacks were the Pb-type defects Pb0 and Pb1 –generally the unique ESR fin-gerprint of an Si/SiO2 interface. Their presence indicated that not only the(100)Si/SiOx/ZrO2 interface, but also the (100)Si/Al2O3/ZrO2 interface isSi/SiO2-type, in terms of inherent interface defects. In the as-depositedsamples the observed signals were weak, but upon VUV activation the de-fect densities exceed those generally observed for standard (100)Si/SiO2.The g matrix appeared to be identical, within experimental error, to thatobserved for the Pb0 and Pb1 defects in SiO2, but the observed signals weresomewhat broadened, indicating that the interface is under enhanced stress.However through subsequent application of an appropriate thermal treat-ment, an interface quality close to that of standard thermal Si/SiO2 couldbe realized.

This was confirmed by electrically detected magnetic resonance on AL-CVD (100)Si/Al2O3 and (100)Si/ZrO2 entities [114] and two X-band ESRworks on ALCVD grown (100)/Al2O3 [115, 116]. Cantin et al. [115] also re-ported on an additional center, the D center, generally ascribed to unpairedSi bonds in disordered a-Si. The authors suggested that part of the defectsis present at the interface or in the dielectric. It is, however, more likelythat all the D centers were located in the lateral cleavage planes and/orsubstrate back. This was suggested by Stesmans et al. [77] and later ondemonstrated by Jones and Barklie [116].

Two more works focussed on the interface defects in Si/HfO2 struc-tures [117, 118]. One work compared (100)Si/HfO2 entities grown by threevariants of CVD [117]. also here, in the as-deposited state, the Pb-type de-fects were reported as predominant defects. In agreement, a second X-bandwork reported the observation of Pb defects at the interface of nominally(111)Si/HfO2(145 nm) entities manufactured via ALCVD using the nitratoprecursor Hf(NO3)4 (NCVD) [118].

One work focussed on the observation of point defects in the SiO2-typeinterlayer in Si/SiOx/ZrO2, Si/Al2O3, Si/Al2O3/ZrO2 and Si/SiOx/HfO2

stacks with nm-thick high-κ layers deposited at 300 ◦C in an ALCVD re-actor [77]. Upon post-deposition oxidation at 650-800 ◦C the E′ and EX


defects –SiO2 associated defects– were observed in all stacks, as well as a95-G doublet in the Si/SiOx/ZrO2 system. It was suggested that the ap-pearance of the SiO2-specific point defects indicated a modification and/oradditional growth of the SiO2-interlayer. The results on the E′ center alsorevealed that the established interlayer in Si/SiOx/ZrO2 is electrically dras-tically inferior to standard thermal SiO2 on Si, while the quality of the interlayer in the Si/HfO2 system was found to be qualitatively close to standardthermal SiO2.

There appears to be a consensus in ESR literature that Pb-type defectsare inevitably present at the Si/high-κ interface, indicating that the for-mation of an SiO2(x) interlayer is endemic for these Si/high-κ metal oxidesystems. This way, as evidenced by both independent electrical measure-ments [108, 109, 110, 111] as well as ESR observations [112, 113, 77], aninterface is realized of technological quality in terms of interface state den-sity and passivation behavior in hydrogen, much like the standard thermalSi/SiO2 system. However the presence of such an interlayer is undesiredin terms of EOT consideration, the more so with increasing relative thick-ness of the ’low-κ’ SiO2 interlayer. Yet, since its occurrence seems hard toprevent in any realistic application, the insertion of an (in thickness) well-controlled SiO2 layer of minimal thickness is currently adopted as a modusvivendi route of progress [94, 119]. The presence of an SiO2(x) interlayer

Figure 1.15: HRTEM images of a Si/SiOx/ZrO2 stack (a) before and (b) afterpost deposition annealing in O2 at 700 ◦C for 15 min.


has more directly been demonstrated by numerous topographic/imagingtechniques, such as medium-energy ion scattering (MEIS), high-resolutiontransmission electron microscopy (HRTEM) and x-ray photoelectron spec-troscopy (XPS) [19, 94, 120, 121, 122, 107, 101]. Fig. 1.15 shows a HRTEMpicture of a Si/SiOx/ZrO2 stack before and after post deposition annealingin O2 at 700 ◦C for 15 min [123]. The 100 nm thin ZrO2 layers were de-posited by ALCVD at 300 ◦C using ZrCl4 and H2O sources. The presenceand additional growth upon annealing (by a factor ∼2.5 in thickness) ofthe SiOx layer is nicely demonstrated.

1.3.2 Intrinsic defects in high-κ oxides

Very little is known about occurring intrinsic point defects in high-κ oxides.Only ZrO2 has been intensively studied as it has been widely used as activecatalyst for many reactions [124, 125, 126, 127]. Three types of defectsare generally observed in ZrO2: (a) an axial symmetric signal generallyattributed to Zr3+ ions in the bulk or at the surface, (b) an isotropic orslightly axial symmetric signal attributed to electrons trapped in oxygenvacancies (F -centers), and (c) a paramagnetic oxygen related center (O−2 -type). However, since the only isotope of Zr with a nuclear spin differentfrom zero has a rather low natural abundance (91Zr; 11.22%) and a highnuclear spin (I=5/2), the observation of any hf splitting originating fromZr may appear quite hopeless. Consequently, the observed ESR signalscannot be conclusively identified and their origin and/or location is stillunder discussion.

For the high-κ oxide currently considered of most technological rele-vance, HfO2, only a few ESR studies, concerning possible intrinsic pointdefects in the dielectric, exist: One work reports on the observation of twoparamagnetic defects in thin ALCVD deposited HfO2 layers on (100)Si af-ter photoinjection of electrons [128]. This is the only work so far existing inthe literature reporting on the ESR observation of intrinsic point defects inhigh-κ layers deposited on Si. Other work reports on the characterizationof defects in monoclinic HfO2 powders [129, 130]. Here a brief overviewwill be given of the different ”intrinsic” point defects observed by ESR inZrO2 and HfO2, with the observed g values gathered in tables 1.5, 1.6, and1.7. To our knowledge no point defects have been observed by ESR in otherhigh-κ oxides than ZrO2 or HfO2.


The Zr3+/Hf3+ defect

In ZrO2 outgassed in vacuum at 623 K and subsequently heated in air,Torralvo et al. observed an axial symmetric ESR signal [124]. Based oncrystal field theory stating that for a d1 ion in a tetragonally or orthorombi-cally distorted cube or in trigonally distorted octahedra the correspondingg values always fulfill g⊥ < ge and g‖ < g⊥, the observed ESR resonancewas assigned to Zr3+ ions. The possibility of an impurity, however, wasalso considered. Later, Azzoni et al. observed a similar axial symmetricsignal in single crystals of yttria stabilized zirconia subjected to x-ray irra-diation [131, 132]. Based on symmetry properties of the observed spectra,concerning g values and line broadening, the authors assigned the defect toa trapped electron in a 4d1 configuration of the 7-fold coordinated zirco-nium atoms near an oxygen vacancy. Morterra et al., however, concludedthat the axial symmetric signal they observed [125] during their study ofthe formation and reactivity of paramagnetic centers in ZrO2 annealed invacuum originates from Zr3+ ions located on the surface of zirconia. Thisassignment was followed by many others [126, 133, 127, 134, 135]. How-ever, Orera et al. did not agree with the surface location of the observedZr3+ defects: Studying intrinsic defects in yttria and calcia stabilized zirco-nia single crystals subjected to x-ray irradiation through combining ESR,optical absorption, and photo emission techniques, the authors assignedthe observed axial symmetric signal to electrons trapped by Zr4+ ions per-turbed by two oxygen vacancies placed at the opposite corners of the anionscube [136]. The idea of the Zr3+ sites located in the bulk of the dielectricnear oxygen vacancies, as originally proposed by Azzoni et al. [131], wasalso adopted by many authors [137, 138, 139, 140, 127].

From the above, it would thus appear there to exist some disagreementin the literature on the location of the Zr3+ sites. It could, however, alsobe that there exist in fact two types of Zr3+ entities exhibiting similarg values, one located at or near the surface and another located in thebulk of the oxide [127]. But it should be kept in mind that without theobservation of any hf structure originating from the low abundant 91Zrnuclei, the assignments of the observed axial symmetric signals to Zr3+

entities is merely suggestive and definitively not conclusive.One other possibility is that the defect is not intrinsic but impurity

related. Ben-Michael et al. observed a sample dependence of the defectdensity of the axial symmetric signal they observed in yttria and calciastabilized zirconia single crystals [141]. Supported by the results of Er-makovich [143] who found a very similar signal in γ-irradiated Ti doped


Table 1.5: Comparison of the different axial symmetric defects observed by ESRin different ZrO2 and HfO2 samples. The list is not exhaustive.

Sample g‖ g⊥ Assignment Ref.ZrO2 gels 1.956 1.981 Zr3+ or impurity [124]

yttria stabilized 1.989 1.852 Zr3+ near [131]ZrO2 O vacancy

ZrO2 powder 1.953-1.975 1.978 Zr3+ at surface [125]yttria and calcia 1.989 1.860 impurity (Ti3+) [141]stabilized ZrO2

yttria and calcia 1.989 1.852 Zr3+ near two [136]stabilized ZrO2 O vacancies

ZrO2 nanopowder 1.961 1.974 Zr3+ near O [137]vacancy in bulk

calcinated zirconium 1.956 1.973 Zr3+ at surface [133]hydroxide

zirconium silicates 1.982 1.973-1.963 Zr3+ inside [127]pore structure

zirconium silicates 1.958 1.977 Zr3+ at surface [127]pure and Cr doped 1.953 1.978 Cr impurity [142]

ZrO2

sulfated ZrO2 1.951 1.979 Zr3+ at surface [126]ZrO2 powder 1.957 1.975 Zr3+ at surface [134](100)Si/HfO2 1.96 1.96 Hf3+ [128]

m-HfO2 powder 1.938-1.941 1.971-1.970 Hf3+ or [129]impurity (Ti3+, Zr3+)

samples, the authors pointed towards the Ti3+ reduced state as being re-sponsible for the defect. This attribution was supported by Merino et al.who found a correlation between the defect density and the Ti concen-tration [144]. Recently, another work [142] reported on the correlation ofthe defect density of the axial symmetric defect observed in pure zirconiaand Cr doped zirconia with the Cr concentration. The assignment wassubstantiated by the observation of the most intense hf component of the53Cr-induced (I=3/2 and 9.501% natural abundant) hf structure.

Also in the HfO2 powders similar axial symmetric signals were ob-served [129, 130]. The authors suggest that the associated defect is probablylocated in the bulk of the material, but are very cautious with the assign-


ment. They suggest that in accordance with the ZrO2 cases reported, thedefect most probably involves Hf3+, but Zr3+ and Ti3+ cannot be ruledout. In (100)Si/HfO2 subjected to a post-deposition N2 anneal at 420 ◦Cand negative charge injection (using the UV light corona ion technique)a weak signal at zero crossing g=1.96 is tentatively assigned to a Hf3+-related ion defect [128] in the HfO2 layer.

It seems that the observation by ESR of the axial symmetric signal inZrO2 and HfO2 (after the appropriate heat treatment or irradiation) isquite systematic. There is however a substantial spread in the reported gvalues for this signal observed in ZrO2: g⊥ ranges from 1.981 to 1.852 andg‖ from 1.989 to 1.951 (see Table 1.5). This inconsistency in the reported gvalues indicates that either the originating defect center must be extremelysensitive to small variations in the ZrO2 matrix leading to the observationof different g values for different ZrO2 samples, or that the observed signalsoriginate from different defect centers. The proposed intrinsic Zr3+ centerlocated in the bulk or at the surface possibly near one ore more oxygen va-cancies and the impurity centers may all be valid possibilities, but withoutthe observation of any hf structure the real origin of these signals remainsobscure.

The F-center

Another signal observed in ZrO2 samples [124, 126, 137, 145, 133, 142,134, 135] and in HfO2 powders [129, 130] exhibits slightly axial or spherical

Table 1.6: Overview of some ESR signals observed in ZrO2 or HfO2 assigned toan F-center.

Sample g‖ g⊥ Ref.ZrO2 gels 2.002 2.004 [124]

ZrO2 nanopowdera 2.003 2.003 [137]calcinated zirconiuma 2.0018 2.0018 [133]

pure and Cr doped ZrO2 2.002 2.002 [142]sulfated ZrO2 2.003 2.003 [126]ZrO2 powdera 2.000 2.000 [134]

m-HfO2 powderb 2.003 2.000-1.994 [129]

a The observed signal was assigned to an F-center located near the surface.b The observed signal exhibited orthorombic symmetry.


Table 1.7: Overview of some ESR signals observed in ZrO2 or HfO2 assigned toan O−2 ion center.

Sample g1 g2 g3 Ref.ZrO2 gels 2.012 2.012 2.004 [124]calcinated zirconium 2.034 2.010 2.010 [133]sulfated ZrO2 2.027 2.010 2.002 [126]ZrO2 powder 2.031 2.009 2.000 [134](100)Si/HfO2 2.04 2.01 2.000 [128]m-HfO2 powder 2.014 2.013 2.006 [129]

symmetry and has a zero crossing g value around g=2.002-2.003, see Ta-ble 1.6. Since this g value is close to the free electron g value (g0=2.0023) ithas been assigned to and F -type defect, i.e., an electron trapped in an O−

oxygen vacancy. Again, also regarding this defect there is some discussionbetween the different authors whether the vacancy is located in the bulk ornear the surface.

The O−2 -type defects

A third defect center observed by ESR in ZrO2 [124, 126, 145, 133, 127, 134]and HfO2 [129, 130, 128] exhibits an othorombic g matrix with one g valueclose to g0 (see Table 1.7). This signal is attributed to an O−2 ion basedon the Kanzig and Cohen model [146] for the g matrix of the O−2 ion inalkali halids widely accepted in literature. From the first order perturbationapproximation of the expressions for the principal g values of an O−2 ion itcan be seen that one g value should be close to g0 and the other two shouldbe greater. The observed O−2 ions in ZrO2 and HfO2 are most probablyadsorbed on the surface.

1.3.3 Impurity related defects in high-κ oxides

It still remains undecided whether the defect centers observed in ZrO2 andHfO2 described in the previous section should be classified as intrinsic orextrinsic point defects. However, up to the current work (chapter 3), theonly undisputable extrinsic defect observed in a high-κ dielectric was theNO2 radical. Stesmans et al. revealed the incorporation of N in NCVD de-posited thin HfO2 films on (100)Si subjected to 60Co γ-irradiation through


the observation of a prominent ESR powder pattern [147]. The spectrumcould not be observed in metallo-organic CVD (MOCVD) nor ALCVDdeposited HfO2 on (100)Si. The spectrum only appeared in the NCVDdeposited samples after γ-irradiation and not after VUV irradiation.

From measurements performed at two different microwave frequenciesthe three observed resonance lines were clearly found to originate from aS=1/2, I=1 (with 100 % natural abundance such as 14N) center. Theprincipal values of the g and hf matrices could be obtained (g1=1.99122,g2=2.00192, and g3=2.00555; A1=49.0 G, A2=68.5 G, and A3=51.0 G)incontestably indicating that the spectrum originates from NO2 radicals(density ≥ 55 at. ppm). The ESR spectrum was observed to remain essen-tially unchanged up to 35 K [148], indicating the NO2 entities to remainimmobilized within the HfO2 matrix. The molecules were found to be sta-bilized and likely homogeneous distributed in the a-HfO2 network. Thedefects precursor was suggested to be the N≡[O–HfO3]3 network entity,the NO2 defects being formed through γ-ray induced structural rearrange-ments.

Chapter 2

Fundamental point defectsin nm-sized silica particles:probing of the networkstructure

2.1 Introduction

That the Si/SiO2 entity is likely to soon lose its position as one of the mainbuilding blocks of MOS entities in semiconductor industry, does not neces-sarily result in a loss of all interest in the SiO2 dielectric. As a result of thecurrent nano-rage the oxide has attracted renewed interest in the form ofnanospheres, nano-agglomerates, nanowires, and nanotubes for their poten-tial use in e.g., opto-electronic nanodevices, catalysis, chromatography, andbiosensors [149, 22, 23]. In this chapter we will focus on flame aerosol man-ufactured silica nanoparticles. Interesting for industry is that these fumedsilica particles can easily be produced in industrial quantities [150]. Conse-quently, from their commercial production in the 1940’s on, the nanoparti-cles were introduced in numerous applications. For instance, the materialcan be found as fillers in toothpaste and car tires, as starting material foroptical fibers it may serve as a general gel former [151], and it is found invarious food such as bread, ice cream, and margarine [152].

But, even though it is well established experimentally that the elec-trical and optical properties associated with fumed silica distinctly dif-fer from the bulk silica counterparts [153, 154, 155], very little is knownabout the exact network structure of the silica particles. One known

63

64 Chapter 2: Fumed silica nanoparticles

piece of information is that the nanoparticles are in the amorphous phase,interpreted as being caused by the extremely fast cooling of the pyro-gene silica aggregates formed at 1400-1800 ◦C. Consequently, recentlya variety of theoretical [156, 157, 158, 159] as well as experimental stud-ies [160, 161, 154, 153, 162, 163, 164, 155, 165, 166] using a range of sensi-tive techniques such as positron annihilation [161], x-ray diffraction [153],and key photon-solid interaction probes including optical [162, 163, 155]and Fourier transform infrared (FTIR) absorption [160, 154, 153], Ra-man spectroscopy [153], nuclear magnetic resonance (NMR) [165], andESR [161, 166], have been reported.

Surprisingly, in the few ESR studies reported, no ESR signals originatingfrom the nanoparticles could be observed [161, 166] which would leavetwo possibilities: The nanoparticles either differ drastically from the bulkcounterparts and to such an extend that the particles would exhibit no pointdefects such as the E′ center, or the occurring (inherent) point defects maybe left passivated by bonding to hydrogen in the as-prepared state, thusrendering them ESR inactive. In view of the particular (H-rich) preparationmethod of fumed silica nanoparticles (vide infra) and based on build upexperience in the Leuven ESR group, the latter seems more likely. Thephenomenon of H-passivation is indeed well known for Si dangling bondtype defects such as the Pb [29, 167, 28] and E′-type defects [168] in thermalSi/SiO2. Irradiation by VUV photons (∼10 eV ) has been proven to be amost suitable approach to dissociate H from passivated defects [62, 63, 169].

Encouraged by the latter idea, ESR research of fumed silica nanoparti-cles was re-opened and indeed, the combination of photon excitation andESR analysis revealed the presence of several (inherent) point defects. Theoccurring ESR resonances are monitored as a function of post-formationheating and treatment (aging, VUV excitation) in this way using the pointdefects as atomic probes of the local network structure. Careful charac-terization and identification of the occurring point defects could possiblyprovide access to the structural nature of the nanoparticles and reveal pos-sible differences as compared to bulk silica. In a next step, inspired byprevious work in our group concerning the structural degradation of ther-mal SiO2 by high-temperature vacuum annealing [63], the possible influenceof SiO on the fumed silica network at elevated temperatures was studied.

2.2 Former studies 65

2.2 Former studies

From a technological as well as a fundamental point of view, it is of much in-terest to obtain a good understanding of the properties of the nanoparticlesvis-a-vis those of the macroscopic counterparts. Here an overview is pre-sented of former studies on fumed silica. Since the nm-sized particles havea very high specific surface area (∼100-400 m2g−1), their surface reactivityand related surface properties have been intensively studied [165, 170, 160].Bunker et al. studied the dissociative chemisorption of various gas specieson the dehydroxylated surface of fumed silica [160]. Using FTIR spec-troscopy their results showed the presence of highly reactive defect sites.The fast hydrolization of these defects, a factor 105 to 5×106 higher ascompared to fused silica, indicates that the reactivity of the Si–O bondsis promoted by bond strain. An extensive NMR research [165] points outthat both mutually hydrogen-bonded and isolated silanols (Si–OH groups)are present at the surface of fumed silica, but only the isolated ones remainafter heating at a temperature Tan>350 ◦C. This work also concludedthat the high temperature production of the nanoparticles leads to a widerrange of variation of Si–O–Si bond angles at the surface of the particles.

Glinka et al. extensively studied the photoluminenscence (PL) proper-ties of 7 and 15 nm sized nanoparticles compared to bulk silica [164, 163].In addition to the PL bands observed for bulk silica –i.e., 1.9 and 2.35 eVPL bands assigned to bulk NBOHCs and hydrogen related species (such as≡Si–H), respectively– an additional PL band was observed at 1.79 eV . Thisextra PL band was attributed to a large concentration of surface NBOHCsin the nanoparticles. They suggested that large concentrations of surfacestructural defects, i.e., E′ (≡Si•) and OHC (≡Si–O•) defects, result from≡Si–O–Si≡ bond splitting during particle formation. The intensity of thePL band attributed to NBOHCs (bulk and surface related) was found toincrease as a function of heat treatment in air (∼2 h) in the range Tan=600-900 ◦C and dominated the spectrum after annealing at Tan∼900 ◦C. Fromthese results the authors concluded that heat treatment of the silica pow-ders in air at Tan≥600 ◦C gives rise to the formation of NBOHCs by thefollowing process:

2(≡ Si−OH) −→ (≡ Si−O•) + (≡ Si•) +H2O ↑ . (2.1)

The concentration of the NBOHCs should thus increase with increas-ing anneal temperature and become dominant for the sample annealed atTan∼900 ◦C.


This picture was partially retained when Altman et al. ascribed the non-exponential behavior of light absorption in the visible range of the fumedsilica nanoparticles to NBOHCs [155]. They suggest that these defects areformed during particle formation. Moreover, the authors believe that theconcentration of these defects may be so high that corresponding strongCoulomb disorder leads to a significant narrowing of the optical band gap.A recent paper [162] reports on white light emission from transparent SiO2

glass prepared from fumed silica, a property that can be of importancefor the field of display and lightning technology. The exact origin of thewhite PL lightening is yet unknown. It was, however, demonstrated thatthe white PL emission cannot arise from the same defects as the previouslyobserved red (∼2.35 eV ) and green (∼1.9 eV ) PL bands originating fromsilica nanoparticles.

Uchino et al. extensively studied the microscopic structure of fumedsilica combining FTIR spectroscopy, Raman spectroscopy, and high-energyx-ray diffraction [154, 153]. In an initial work they studied the modifica-tion of the fumed silica structure under pressure [154]. Compressing thenanoparticles at room temperature at 2 and 5.5 GPa resulted in opaqueand translucent samples. At higher pressures (6 and 8 GPa) transparentSiO2 glass was obtained structurally different from normal a-SiO2. Thisirreversible pressure-induced structural transition occurs at lower pressures(2-8 GPa) than would be expected for bulk silica glass (>10 GPa). As

0 200 400 600 800 1000 1200

0

100

200

300

400

BE

T su

rface

are

a (m

2 /g)

Tan (°C)

Figure 2.1: BET surface areas of fumed silica samples heated in air at differenttemperatures. (Data taken from Ref. [153].)

2.2 Former studies 67

these results are most likely linked with the intrinsic structural charac-teristics of the nanoparticles, this stimulated the authors to extend theirstructural analysis of fumed silica [153].

Looking at the Brunauer-Emmet-Teller (BET) specific surface areas ofthe as-recieved and heat-treated (∼2 h in air) fumed silica particles Uchinoet al. found that the surface area decreases upon heating as illustratedin Fig. 2.1. This decrease was attributed to sintering of the nanoparticleswith an onset temperature of ∼850 ◦C, as proposed previously [171]. Uponheating at ∼1200 ◦C the surface area dropped below the detection limit(∼1 m2g−1). This was explained as resulting from the coalescence of theprimary particles into bulk-like silica. These changes in the particles mor-phology could also nicely be seen from the field emission scanning electronmicroscopy (FESEM) pictures shown in Fig. 2.2.

The FTIR spectra of the fumed silica particles heated up to ∼1100 ◦Cdistinctly differ from those of the sample heated at ∼1200 ◦C which wasvery similar to that of bulk silica, as can be seen from Fig. 2.3. Theirresults thus indicated that the nanoparticles in the as-received state orafter annealing in the range 900-1100 ◦C are structurally different frombulk silica while the fumed silica sample heated at ∼1200 ◦C becomes verybulk silica like. In agreement with the FTIR spectra, the fourier trans-

sintering is;850 °C regardless of particle characteristics orheating times.28 If we assume that each fumed silica particleis ideally spherical, we can estimate average diameter~d! ofprimary particles using a relationshipd56.03103/(rS)wherer is the specific density of primary particles of fumedsilica ~2.2 g/cm3!, S in m2/g, andd in nm. According to thisrelationship,d is estimated to be 13 and 24 nm at 1000 and1100 °C, respectively. Considering thatd of the as receivedsample is 7 nm, we suggest that two or three particles ap-proach one another and induce sintering or consolidation toform slightly larger primary particles in the temperaturerange 1000–1100 °C. FESEM observations also reveal thedevelopment of particle size with temperature~see Fig. 1!.We should note, however, that the fumed silica samplesheated at 1100 °C still retain the morphology of the originalnanometer-sized silica particles. It is probable that thepresent sintering process does not occur by melting sincemelting of silica would require temperatures in excess of2000 °C, implying that the present consolidation is inducedby viscous sintering.

We have found thatSdecreases to the detection limit~;1m2/g! of the present volumetric adsorption apparatus at1200 °C, which is near the glass transition temperature ofsilica glass.29 This indicates that fine primary particles coa-lesce into bulklike silica at 1200 °C. Indeed, the sintered ma-terial heated at 1200 °C was not aggregates of nanometer-sized fine powders, as elucidated by the FESEM imageshown in Fig. 1~d!, but a collection of relatively large~;0.5mm! grains.

B. Infrared absorption spectra

Figure 2 shows FTIR spectra of heat-treated fumed silicasas a function of the temperature of dehydroxylation alongwith an FTIR absorption spectrum of bulk silica glass. Inagreement with the previous experiments, the infrared spec-trum of bulk silica glass exhibits three main absorption bandsat ;1100, ;800, and;480 cm21, which are assigned toasymmetric stretching, symmetric stretching and bendingmotions of the SiuOuSi bonds, respectively.30,31The FTIRspectrum of fumed silica heated at 1200 °C is almost compa-rable to that of the bulk silica glass. This indicates that sin-

tering is almost completed at 1200 °C to form coalescentbulklike materials, in accordance with the results mentionedin Sec. III A.

It should be noted, however, that FTIR spectra of fumedsilica samples heated below 1100 °C have different spectralfeatures than the one heated at 1200 °C or that of bulk silicaglass. One of the striking differences can be seen in theSiuOuSi asymmetric stretching band at;1100 cm21; inthe FTIR spectra of fumed silica heated at 900–1100 °C thisband has rather a narrow feature as compared with the cor-responding band of bulk silica glass. In addition, theSiuOuSi symmetric stretching band at;800 cm21 ob-served for fumed silica samples heated at 900–1100 °C islocated slightly at higher frequencies~808 cm21! than that ofthe bulk~800 cm21!. These results suggest that the structureof dehydroxylated fumed silica is not identical to the struc-ture of bulk silica glass; that is, the size dependence is seenin the structure of silica glass.

Figure 3 shows the infrared absorption spectra of an over-tone mode of the SiuOuSi stretching band for the heat-treated fumed silica samples. We see from Fig. 3 that thepeak position of this overtone mode shifts to higher frequen-cies with increasing heat-treated temperature. According tothe empirical relationship between the peak position of theovertone moden and the fictive temperatureTf of silicaglass20

n52228.641~43809.21/Tf ! ~1!

we estimated the fictive temperature of the present fumedsilica samples~see Table II!. We see from Table II that thefictive temperature of the fumed silica decreases with in-creasing heating temperature. We should note, however, thatthe values of fictive temperature for the samples heated at

FIG. 1. FESEM images of fumed silica after heat treatment at~a! 900, ~b! 1000,~c! 1100, and~d! 1200 °C for 2 h.

FIG. 2. Fourier transform infrared~FTIR! absorption spectra offumed silica after heat treatment between 900 and 1200 °C. Anadditional weak shoulder at;980 cm21 is attributed to the bendingvibration of residual SiuOH groups~Ref. 46!. FTIR spectrum ofnormal bulk silica glass is also shown.

MICROSCOPIC STRUCTURE OF NANOMETER-SIZED . . . PHYSICAL REVIEW B69, 155409 ~2004!

155409-3

Figure 2.2: FESEM images of fumed silica after heat treatment in air for 2 h at(a) 900 ◦C, (b) 1000 ◦C, (c) 1100 ◦C, and (d) 1200 ◦C (taken from Ref. [153]).


form Raman spectra of the as-received sample and samples heat treated atTan∼900-1100 ◦C differ from bulk silica, but nicely coalesce when heatedat ∼1200 ◦C (see Fig. 2.4). The bands at 495 and 606 cm−1 attributed tofour and three membered silica rings, respectively, were found to be morepronounced for the as-received particles than for bulk silica indicating thatthe population of small membered rings in the nm-sized silica particles islarger than in bulk silica. From their in situ infrared absorption spectrathe authors concluded that the fumed silica network is more flexible andmore rigid against applied pressure than bulk silica.

The fumed nanoparticles did not only attract the attention of experi-mentalists, but were also studied theoretically [156, 157, 158, 159]. Severalmolecular dynamics (MD) simulations have been reported for silica clusterscontaining ∼103 to ∼104 atoms, revealing specific structural properties forthe nm-sized particles [159, 156, 157]. From the simulations it was con-cluded that the nanoparticles have a shell-like structure: The structure anddensity of the surface of the nanoparticles are substantially different fromthose of the interior of the cluster. The interior of the cluster was suggestedto be structurally equivalent to bulk silica. These results, however, are notcompletely consistent with the experimental results obtained by Uchino etal. [153] as they concluded from their measurements that the structure ofthe interior of the nanoparticles cannot be equivalent to that of macroscopicSiO2.

It has been demonstrated that inherent structural defects, as studied bythe ESR technique, may serve as atomic sized probes of utmost sensitivityto structural aspects of their local environment (see, e.g., Refs. [172, 29,173]). So, their characterization and identification may appear a key meansto assess the structural nature of the nanoparticles and to reveal possibledifferences as compared to the bulk silica counterpart. Though being atool with atomic level physico-chemical sensitivity, very little research hasso far been carried out using ESR: Some ESR work has been done on fumedsilica in the as-grown state [166] as well as after UV irradiation [161] withphotons (∼5 eV ) obtained from a low pressure Hg lamp. However, asmentioned before, in both cases no ESR signals originating from the silicananoparticles could be observed, likely explaining the faded interest overthe years in applying ESR to this field of research.

2.3 Interaction with SiO 69

sintering is;850 °C regardless of particle characteristics orheating times.28 If we assume that each fumed silica particleis ideally spherical, we can estimate average diameter~d! ofprimary particles using a relationshipd56.03103/(rS)wherer is the specific density of primary particles of fumedsilica ~2.2 g/cm3!, S in m2/g, andd in nm. According to thisrelationship,d is estimated to be 13 and 24 nm at 1000 and1100 °C, respectively. Considering thatd of the as receivedsample is 7 nm, we suggest that two or three particles ap-proach one another and induce sintering or consolidation toform slightly larger primary particles in the temperaturerange 1000–1100 °C. FESEM observations also reveal thedevelopment of particle size with temperature~see Fig. 1!.We should note, however, that the fumed silica samplesheated at 1100 °C still retain the morphology of the originalnanometer-sized silica particles. It is probable that thepresent sintering process does not occur by melting sincemelting of silica would require temperatures in excess of2000 °C, implying that the present consolidation is inducedby viscous sintering.

We have found thatSdecreases to the detection limit~;1m2/g! of the present volumetric adsorption apparatus at1200 °C, which is near the glass transition temperature ofsilica glass.29 This indicates that fine primary particles coa-lesce into bulklike silica at 1200 °C. Indeed, the sintered ma-terial heated at 1200 °C was not aggregates of nanometer-sized fine powders, as elucidated by the FESEM imageshown in Fig. 1~d!, but a collection of relatively large~;0.5mm! grains.

B. Infrared absorption spectra

Figure 2 shows FTIR spectra of heat-treated fumed silicasas a function of the temperature of dehydroxylation alongwith an FTIR absorption spectrum of bulk silica glass. Inagreement with the previous experiments, the infrared spec-trum of bulk silica glass exhibits three main absorption bandsat ;1100, ;800, and;480 cm21, which are assigned toasymmetric stretching, symmetric stretching and bendingmotions of the SiuOuSi bonds, respectively.30,31The FTIRspectrum of fumed silica heated at 1200 °C is almost compa-rable to that of the bulk silica glass. This indicates that sin-

tering is almost completed at 1200 °C to form coalescentbulklike materials, in accordance with the results mentionedin Sec. III A.

It should be noted, however, that FTIR spectra of fumedsilica samples heated below 1100 °C have different spectralfeatures than the one heated at 1200 °C or that of bulk silicaglass. One of the striking differences can be seen in theSiuOuSi asymmetric stretching band at;1100 cm21; inthe FTIR spectra of fumed silica heated at 900–1100 °C thisband has rather a narrow feature as compared with the cor-responding band of bulk silica glass. In addition, theSiuOuSi symmetric stretching band at;800 cm21 ob-served for fumed silica samples heated at 900–1100 °C islocated slightly at higher frequencies~808 cm21! than that ofthe bulk~800 cm21!. These results suggest that the structureof dehydroxylated fumed silica is not identical to the struc-ture of bulk silica glass; that is, the size dependence is seenin the structure of silica glass.

Figure 3 shows the infrared absorption spectra of an over-tone mode of the SiuOuSi stretching band for the heat-treated fumed silica samples. We see from Fig. 3 that thepeak position of this overtone mode shifts to higher frequen-cies with increasing heat-treated temperature. According tothe empirical relationship between the peak position of theovertone moden and the fictive temperatureTf of silicaglass20

n52228.641~43809.21/Tf ! ~1!

we estimated the fictive temperature of the present fumedsilica samples~see Table II!. We see from Table II that thefictive temperature of the fumed silica decreases with in-creasing heating temperature. We should note, however, thatthe values of fictive temperature for the samples heated at

FIG. 1. FESEM images of fumed silica after heat treatment at~a! 900, ~b! 1000,~c! 1100, and~d! 1200 °C for 2 h.

FIG. 2. Fourier transform infrared~FTIR! absorption spectra offumed silica after heat treatment between 900 and 1200 °C. Anadditional weak shoulder at;980 cm21 is attributed to the bendingvibration of residual SiuOH groups~Ref. 46!. FTIR spectrum ofnormal bulk silica glass is also shown.


155409-3

Figure 2.3: FTIR absorption spectra of fumed silica after heat treatment in airfor 2 h in the range Tan∼900-1200 ◦C together with the FTIR spectrum of normalbulk silica glass. All spectra are taken from Ref. [153].

bending motions of oxygen atoms in the random network.These two sharp bands at 495 and 606 cm21 are referred toasD1 andD2 , respectively, and are attributed to symmetricoxygen ring breathing vibrations of regular four-membered(D1) and three-membered (D2) silica rings.36–38 As shownin Fig. 5, theD1 andD2 bands can be seen in the FT-Ramanspectra of the heat-treated fumed silica samples as well. It isclear that the intensities of theD1 and D2 bands are ratherhigh for the sample heated at 900 °C and tend to decreasewith increasing temperature. When the temperature reaches1200 °C, the FT-Raman spectrum of the heat-treated fumedsilica becomes closely analogous to that of bulk silica glass,in agreement with the measurements of surface areas,FESEM, and infrared spectra mentioned earlier. It is hencequite likely that the concentrations of theD1- andD2-relatedstructural units, or four-membered and three-memberedsilica rings, respectively, are dependent on the size of theprimary particles. That is, in nanometer-sized silica particlesthe population of small-membered rings is larger than theone in the bulk. These ‘‘regular’’ rings are transformed intolarger ‘‘disordered’’ rings with increasing temperature, lead-ing to the atomic configurations similar to those in the bulkstructure.

Previous Raman studies of silica glass versusTf havedemonstrated that the intensities of theD1 and D2 bandsdecreases with decreasingTf .22 This result is also in quali-tatively agreement with the temperature dependence ofTfobtained from FTIR measurements shown in Table II.

D. X-ray diffraction measurements

The total Faber-Ziman structure factors39 S(Q), for thebulk silica glass and the heat-treated fumed silica samplesare shown in Fig. 6. As far as theS(Q) data are concerned,we do not see any noticeable difference between the heat-treated fumed silica and the bulk silica glass except for theQ

range below;0.5 Å21. An abrupt increase seen inS(Q) ofheat-treated fumed silicas probably results from the struc-tural fluctuation or the aggregated structure formed from thenanometer-sized primary particles.13 To get the informationin real space, we calculate total correlation functions,T(r ),~see Fig. 7! from the weighted interference functions

FIG. 5. Fourier transform Raman~FT-Raman! spectra of fumedsilica after heat treatments between 900 and 1200 °C. FT-Ramanspectrum of normal bulk silica glass is also shown.

FIG. 6. The x-ray structure factorsS(Q) of normal bulk silicaglass and heat-treated fumed silica. Successive curves are displacedupward by 1 for clarity.

FIG. 7. ~a! Total correlation functionsT(r ) of normal bulk silicaglass and heat-treated fumed silica.~b! Expand plot ofT(r ) in thedistance region from 2.8 to 4.2 Å. The curve shown in the bottom isobtained by subtractingT(r ) at 1000 °C from the one at 1200 °C.


155409-5

Figure 2.4: Fourier transform Raman spectra of fumed silica after heat treatmentin air for 2 h in the range Tan∼900-1200 ◦C together with the Fourier transformRaman spectrum of normal bulk silica glass. All spectra are taken from Ref. [153].


2.3 Interactions with SiO at elevated tempera-tures

In previous work on the structure of thermal oxide on Si [63], the influenceof post-deposition heat treatment in an O2-deficient ambient in the tem-perature range Tan=950-1250 ◦C was studied. The main motive for thiswork was probing by ESR via generated point defects of the electricallyand morphologically well known thermal degradation process effectuatedby such annealings to the thermal Si oxide on Si (see, e.g., Refs [174, 62]).In another ESR work the interaction of fused silica with gaseous SiO, insitu provided from co-inserted solid SiO, during annealing at ∼1140 ◦C wasstudied [65]. One of the incentives for these studies was the suspicion raisedabout the influence of Si/SiO2 interface-released SiO molecules: Indeed pre-vious work had observed the formation of pores (voids) through the SiO2

film evidencing the freeing of volatile SiO at the Si/SiO2 interface throughthe net reduction reaction Si (s) + SiO2 (s) → 2SiO (volatile) [174]. Later,

1#. However, it was observed that before VUV irradiation,when these additional signals are not detected, theS signalexhibits approximately the same characteristic weakly asym-metric signature.

Through optimization of the spectrometer settings and in-tense signal averaging, much effort was given to detect ad-ditional signals that could be attributed to hyperfine~hf!structure. Reproducible structures, albeit with limited resolu-tion and obscured by interfering background signals, wereobserved at distances larger than;40 G from the centralZeeman line. The assignment will be discussed below.

The ‘‘distorting’’ [email protected].,~B!# appear to be saturatedsignals mainly due toEg8 .27 This is demonstrated in the insetof Fig. 1, closing in on theg'1.997– 2.005 window. Heresaturation was avoided by reducingPm to 4–8 nW. Severalsignals can be recognized. TheEg8 signal displays a distinctpowder pattern~PP! line shape and is most dominant@marked ~B! in Fig. 1#. At g52.0019 an additional weaknarrow line is observed@~C! in Fig. 1#, also previously ob-served and attributed to theEd8 defect. Finally, at g52.0025 another small narrow signal@~D!# appears repro-ducibly, which in line with previous observations is attrib-uted to theEX center. It needs to be remarked, however, thatthe presence ofEX in the presently studied series of samplesis somewhat different from the previous observation.6 In thelatter study the distinct simultaneous presence ofEX and Sdefects was observed~cf. Fig. 1 in Ref. 6!, which, in thepresent study, was not observed.

Figure 2 shows the inferred areal densities of the observeddefectsS, Eg8 , Ed8 , andEX as a function ofTan. The quoteddensities for all signals, includingS, were obtained assumingspin s51/2 defects and Curie-type magnetic susceptibility.The densities were measured both before and after VUV ir-radiation. In the former case only theS signal could be dis-cerned, appearing only forTan>1050 °C. From this tempera-ture on, the density ofS ~solid squares in Fig. 2! increasesmonotonically with increasingTan up to 1200 °C. After thelast Tan increment (Tan51250 °C), an additional large in-crease (factor;10) in @S# is observed, with some oxide de-

generation also becoming visible optically. The distinct im-pact of the Tan51250 °C POVA step was also exposedelectrically. Unlike all lowerTan samples~< 1200 °C!, it wasnoticed during VUV irradiation that the leakage currentthrough the oxide upon applying the standard120 V wasvery high ~;1 mA/slice!, which forced reduction of the bi-asing voltage to110 V. BelowTan51050 °C, weak traces ofthe signal nearg52.0027 could only be observed in theTan51000 °C sample. Possibly these traces are also due toS,but competition with the noise makes the integration trouble-some. A rough estimate of the signal intensity would corre-spond to an areal density below 131012cm22.

After VUV irradiation, the qualitative behavior of@S#with Tan is approximately the same. However, an overall

FIG. 2. Areal defect densities ofS ~h!, Eg8 ~L!, Ed8 ~n!, andEX~s! defects inferred fromK-band ESR in standard thermal Si/SiO2

after ;1 h vacuum anneal at various temperatures, before~solidsymbols! and after VUV irradiation under positive bias~standardly120 V!. Dashed lines represent best fits of Eq.~9!, from where theindicated activation energies (Ea) are inferred.

FIG. 1. K-band ESR spectra observed at 4.2 Kin standard thermal Si/SiO2 after ;1 h vacuumannealing at various temperaturesTan. The leftpart exposes the growth with increasingTan of abroad signal~A! attributed to theS center. Theright part ~inset!, magnifying the g;1.997–2.005 region, was measured at lower microwavepower to avoid saturation, thus correctly expos-ing the additional narrow signals, remaining par-tially saturated in the left side spectra. The signals~B! and ~C! originate fromEg8 and Ed8 , respec-tively. Additionally, reproducible traces of theEXsignal are observed~D!. The1 symbols representa best Lorentzian fit to the largest intensitySsig-nal. s symbols represent a tentative ‘‘powderpattern’’ fit assuming an axial dangling bond typedefect~see text!. The signal atg51.998 69 stemsfrom a comounted Si:P marker.

STRUCTURAL DEGRADATION OF THERMAL SiO2 ON . . . PHYSICAL REVIEW B 66, 045307 ~2002!

045307-3

Figure 2.5: Areal defect densities of S(�), E′γ(♦), E′δ(4), and EX(©) defectsinferred from K-band ESR in standard thermal Si/SiO2 after ∼1 h vacuum annealat various temperatures, before (solid symbols) and after VUV irradiation underpositive bias (standardly +20 V ). The figure was taken from Ref. [113].

2.4 Experimental details 71

the process of outdiffusion and subsequent desorption of the volatile SiOfrom thermal Si/SiO2 was evidenced more directly by x-ray photo-electronspectroscopy during post-oxidation vacuum annealing at 800 ◦C [175, 176].Nakamura et al. [177, 178] demonstrated, using multi-photon ionizationspectroscopy, the formation of SiO molecules by the reaction of a Si surfacewith O2: O2+2Si→2SiO. The question remained whether the outdiffusionof gaseous SiO would influence the SiO2 network. It was found that theSiO did indeed affect the SiO2 network, in fact aggressively degrading it.Such information was revealed through the observation of several types ofgenerated defects including E′γ , E′δ, EX, and the predominant exclusive Scenter. The areal defect densities of the occurring point defects as a func-tion of the in vacuum (∼1 h) annealing temperature are shown in Fig. 2.5.A specific finding was that the interaction of the SiO2 network with SiO atelevated temperatures resulted in an increase in E′γ density.

2.4 Experimental details

2.4.1 Samples

The samples studied were taken from high-purity pyrogene fumed silicapowder of 7 nm average particle size and 380±40 m2g−1 surface area, witha low metallic impurity content. The samples were obtained from Sigma-Aldrich Inc., Missouri, USA. The structure of the material is amorphous.During particle formation and subsequent cooling down, interparticle col-lisions and subsequent fusion results in the formation of chain-like aggre-gates from 10 to 30 units, or, put differently, from ∼0.1 to 0.2 µm inlength. Since fumed silica is fabricated by burning silicon tetrachloridein an oxygen-hydrogen flame at ∼1800 ◦C, nominally there should be nocarbon present. During formation of the product, hydroxyl groups ( 3.5-4.5×103 µm−1) become attached to some of the silicon atoms on the surfaceof the silica particles.

2.4.2 ESR spectrometry

Conventional cw absorption-derivative ESR measurements were performedat X (∼9.2 GHz), K (∼20.4 GHz), and Q-band (∼33 GHz) in the tem-perature range 4.2-300 K. The levels of the magnetic field ( ~B) modulationBm and incident microwave power Pµ were properly watched to avoid sig-nal distortion. The hf signals were measured in an (out-of-phase) secondharmonic mode in Q-band at relatively high modulation amplitudes (2 G)


and Pµ (5 mW ). Typically, an ESR sample comprised ∼3 mg of fumedpowder.

Separate sets of samples (fresh ones for each thermal step) were sub-jected to post-manufacture baking in vacuum (base pressure <4×10−6

Torr) for ∼1 h at desired temperatures (Tan) in the range 850-1115 ◦C.To study the interaction with SiO at elevated temperatures, again separatesamples were subjected to post-manufacture baking in vacuum (base pres-sure <4×10−6 Torr) for ∼3 h at desired temperatures (Tan) in the range1005-1205 ◦C. But this time each heating procedure was carried out witha suitable amount of the powder tightly sandwiched between two (fresh)slices of Si wafer (0.47 nm thick) with a standard thermal SiO2 (6.5 nm)on top, the SiO2 film facing the inside of the sandwich, as schematicallypictured in Fig. 2.6. In this way, it was figured that in diffusing out theSiO released at the Si/SiO2 interface of the slices could interact with thesandwiched fumed particles.

From previous investigations it may be anticipated that in the as-receivedfumed silica particles, the particular pyrolytic fabrication technique applied(oxy-hydrogen flame) may have resulted in passivation of dangling-bondtype point defects by hydrogen (pictured as≡SiH,≡Si–OH formation), thusleaving possibly occurring inherent systems of defects in the diamagneticstate and hence disabling ESR detection. More specifically, for defects suchas the E′γ centers, occurring (passivated) precursor sites are pictured as theO3≡Si–H entity at the site of an O vacancy [48, 179], while, similarly, forthe NBOHC, O3≡Si–OH (hydroxyl incorporation into the silica matrix) is

SiO release

SiO release

Figure 2.6: Sketch of the Si/SiO2/fumed silica/SiO2/Si sandwich used for hightemperature (1005-1205 ◦C) SiO–fumed silica interaction treatment.

2.5 Probing of point defects 73

pictured as an obvious site [48]. So, after initial ESR tests, to maximally re-veal defects, samples (both in the as-received state and after additional heattreatment) were subjected at room temperature to prolonged irradiation byVUV (10 eV ) photons (flux ∼1015 cm−2s−1) obtained from a Kr-resonantdischarge lamp sealed with a MgF2 window to photodissociate H from pas-sivated defects, or put differently, to radiolyse defect–H bonds. Possibly, thetreatment may additionally unveil strained or weak bondings (bond rup-ture) and diamagnetic precursor sites [62, 63]. This, however, would alsoadd to the ultimate goal, that is, acquiring atomic scale information aboutthe SiO2 particles network, as these sites of defect creation, i.e., strainedand/or weak bonding sites, constitute in embryo also imperfections in thea-SiO2 matrix, although not necessarily concerning nonstoichiometry.

Early on in this work aging effects have been observed to occur in roomambient for at least one defect (vide infra). So ESR measurements werefirst performed immediately upon VUV irradiation and subsequently re-peated after the samples were left in room ambient for desired times (oneday to weeks). This enabled us to study the characteristics of defect modifi-cation in room ambient. For reference, measurements were also carried outon a suprasil I synthetic silica sample, obtained from Heraeus (Germany)containing [OH]∼1200 ppm (by weight), [H2O]≤600 ppm, and total metalcontent <1 ppm, subjected to 100 Mrad(Si) 60Co γ-irradiation at 300 K.

2.5 ESR probing of occurring point defects

As anticipated, no ESR signal could be observed in the as-fabricated silicaparticles. This is ascribed, at least in part, to defect inactivation by hydro-gen due to the abundance of H2 during the flame growth. But that situationchanges drastically upon VUV irradiation or after annealing at high tem-perature. Several signals are observed, their proper spectroscopic isolation,though, being strongly hampered by signal overlap and entanglement. Asa result, while at least 5 types of signals could be assigned and analyzed,several interesting resonant responses (mostly of low intensity) remain asyet unidentified. Table 2.1 gives an overview of the observed defect densi-ties as a function of post-manufacture heating and VUV treatment. Thevarious types of defects observed will now be addressed separately.

The presence of SiO release during vacuum annealing at high temper-atures only had a significant influence on the observations concerning theE′-type centers. Consequently, the effects of the vacuum anneal in thepresence of SiO release (henceforth referred to as SiO-vac. anneal) will be


Table 2.1: Overview of typical densities of the observed defects as a function ofthe anneal temperature Tan (vacuum anneal) and VUV treatment in fumed silicapowder. Quoted values are in units of 1014 g−1. To convert the densities to units of1014 cm−3, multiply by 2.3. Entries left blank mean that the corresponding defectcould not be detected or the signal appeared too weak for reliable discrimination.

Sample condition [E′] [LU2] [OHC] [EX] [CH3]As-receivedAs-received + VUV [a] 2.2±0.5 31±3 0.08±0.01Tan=850 ◦C 0.52±0.05Tan=850 ◦C + VUV [a] 2.4±0.5 0.47±0.05Tan=850 ◦C + VUV [b] 1.0±0.5Tan=960 ◦C 60±6 2±1Tan=960 ◦C + VUV [a] 1.5±0.5 7.3±0.8 2±1 0.14±0.02Tan=1005 ◦C 7±0.5Tan=1005 ◦C + VUV [a] 3±1 1.5±0.2 0.07±0.01Tan=1115 ◦C 49±5Tan=1115 ◦C + VUV [a] 3.5±1 21±3

[a] The densities given here are obtained from samples subjected to ∼1 h VUVirradiation.[b] Several days after VUV irradiation, the sample being left in room ambient.

fully included in the following subsection dealing with the observed E′-typedefects.

2.5.1 The E ′-type center

Observations

The effect of VUV treatment and heat treatment in vacuum

After VUV irradiation, the SiO2-specific E′ type center (generic entityO3≡Si•) is clearly detected in the as-received sample, as shown in Fig. 2.7for X-band and in Fig. 2.8 (a) for K-band observations, as well as in thesamples subjected to post-manufacture baking. A characteristic central-line two-peak powder pattern signal is observed much similar (e.g., re-garding width, overall shape) to that of the common E′γ variant [180, 46](see section 1.2.2) in bulk silica, characterized by [4] the principal g ma-trix values g1=2.0018, g2=2.0006, and g3=2.0003. In the K-band ESRspectroscopy a zero crossing g value of gc=2.00058±0.00005 is typicallyobserved [181, 182] in thermally grown SiO2 on Si. Detailed computer


3276 3278 3280 3282 3284 3286 3288

E'γ

as-received + VUV9.17 GHz; 110 K

Si:P

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

g=1.99891

2.0007

Figure 2.7: Powder pattern derivative-absorption X-band ESR spectrum of theE′ signal measured immediate upon VUV activation of the as-received fumed silicananoparticles, using Pµ=0.1 mW , Bm= 0.6 G, and averaging over ∼50 scans.

simulations of the presently observed spectra show slight, but convincing,variations in ESR parameters depending on the applied anneal tempera-ture, elapsed time period between VUV irradiation and ESR measurementas the sample was kept in room ambient, and the sample temperature dur-ing the measurements.

In the as-received sample, measured immediately after VUV irradiation,the observed zero-crossing g value gc=2.00076±0.00005 is slightly, yet welloutside experimental error, larger than the one observed for irradiated ther-mally grown SiO2 on Si or fused silica, i.e., gc=2.00058±0.00005 [46, 181,182]. As a reference the K-band spectrum for suprasil I synthetic silica sub-jected to 100 Mrad(Si) 60Co γ-irradiation at 300 K is shown in Fig. 2.8(c). Additional heat treatment, however, affects the E′ signal distinctly.Upon annealing at increasingly higher temperatures in the range Tan>850◦C, the zero-crossing g value gradually decreases for Tan→1115 ◦C towardsthe value typically measured for E′γ in bulk silica’s. There is one more re-markable observation: Pertinently, when, after VUV irradiation, leaving


2.00058(c) fused silica+ γ irrad.

as-grown + VUV (adapted)(b)

(a) as-grown + VUV (immediate)

Si:P20.65 GHz; 4.2 K

2.00054

2.00076 1.99869

5G

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

Figure 2.8: Powder pattern derivative-absorption K-band ESR spectra (Pµ∼0.05-0.2 nW; Bm=0.35-0.5 G) observed at 4.2 K of the E′ centers in fumed silica (a)immediate upon VUV irradiation, (b) several days later the sample being left inroom ambient, and (c) of the E′γ centers in suprasil I synthetic silica subjected to100 Mrad(Si) 60Co γ-irradiation at 300 K. Dashed lines represent spectra simula-tions employing Gaussian broadening functions. The used g values and correspond-ing peak-to-peak derivative line widths are (a) g1=2.00174; 2.1 G, g2=2.00070; 2.3G, g3=2.00030; 2.2 G; (b) g1=2.00174; 0.9 G, g2=2.00048; 1.4 G, g3=2.00017;2.1 G; (c) g1=2.00175; 0.80 G, g2=2.00057; 1.45 G, g3=2.00032; 1.30 G.


0 200 400 600 800 1000 1200

2.00052

2.00056

2.00060

2.00064

2.00068

2.00072

2.00076

2.00080

g c

Tan (°C)

Figure 2.9: Zero-crossing g value of the measured E′ signal (K-band; 4.2 K) asa function of post-manufacture heating in vacuum. Closed red symbols: measuredimmediate upon VUV irradiation. Open symbols: several days later, the samplesbeing left in room ambient. The red curve is merely meant to guide the eye.

the samples for substantial time (∼days) in room ambient –henceforth, re-ferred to as the adapted state– the zero-crossing g value has shifted aboutthe same value (towards gc=2.00054±0.00005) for all the samples regardlessof their heat treatment including the as-grown sample [cf. Figs. 2.8 (a) and(b)]. These trends are exposed in Fig. 2.9, where gc measured immediateupon VUV irradiation and after subsequent stay of the samples for days inroom ambient, is plotted against Tan.

A second important ESR parameter is the displayed powder pattern lineshape. A likely pertinent observation is that the shift in gc over time in roomambient is accompanied by an attendant variation in the displayed powderpattern line shape. As illustrated in Fig. 2.8 (b) for the as-grown fumedparticles, this implies, as general trend, evolution from a more blurred,broadened shape to a sharper, well expressed typical ”two-peak” powderpattern pertaining to a nearly axially symmetric g matrix. A similar trendis observed as the measurements are carried out at different temperatures:The lower the sample temperature, the more blurred and broadened thepowder pattern becomes, as illustrated in Fig. 2.10. At the same time asmall variation in gc is observed. Careful analysis leads to the conclusionthat the latter variation in zero crossing g value is purely a consequence


7395 7400 7405 7410 7415 7420

Si:PE' 2.00067

as-grown + VUV20.74 GHz

77 K

26.7 K

14.5 K

4.2 KdP

µ/dB

(arb

. uni

ts)

magnetic field (G)

Figure 2.10: Powder pattern K-band (Pµ∼2.5 nW , Bm∼0.6 G) ESR spectra of theE′ signal measured at different temperatures in the as-received sample immediateupon VUV irradiation. It should be noted that, as known, g(Si:P) is temperaturedependent [16].

of the attendant strong change in the line shape and has little to do withthe gc shifts mentioned before. As will be outlined later, it is attributed toT -dependent surface strain adaptation.

A third ESR parameter able to provide important information is thedefect (E′) density. Immediately upon prolonged (days-weeks) VUV ac-tivation of the as-grown particles, the maximum density that could beattained was determined as (2.4±0.5)×1015 g−1, corresponding to about1×10−3 defects/nanoparticle. It should be added though that this numbermay not represent the maximum number of E′ type centers inherent tothe fumed nanoparticles, as the VUV treatment may not have exhaustivelyactivated all defect sites present due to the small penetration depth (∼10nm) of the VUV photons in SiO2 in combination with the powder nature


of the sample (geometric shadowing). However, the number may be close,given the prolonged irradiation treatment with regular intermittent stirringof the fumed powder. In support, doubling the irradiation time, indeed,did not increase any further the number of defects detected.

At this junction, it may be instructive to compare the observed E′ densi-ties with values encountered in the other macroscopic silica (silica glasses).Obviously, such comparison should be carried through most relevantly fordefects produced by similar VUV treatment. A first most representativebenchmark here may appear dry thermal SiO2 grown on c-Si substrates,ubiquitously used in semiconductor IC technology and known to be of su-perb electrical quality, viz., in terms of point defect-related charge trappingcenters. Remarkably, it is found that the measured E′ density is about 100times lower [183] than the densities typically observed [182, 181] (∼1.4×1017

g−1) in thin (∼4.1-6.5 nm thick) thermally grown dry SiO2 on Si, indicatingthat the fumed silica is of superb quality in terms of malignant oxygen va-cancies, that is, close to perfect SiO2 stiochiometry or O-enriched. Further,it may be appropriate to add that the E′-type signal so far reported forthermal SiO2 seems to be similar to the familiar E′γ signal found in bulksilica. However, no detailed study on the hf structure of the E′γ centersobserved in thermal SiO2 appears available yet, without doubt partiallybecause of limited ESR sensitivity. Second, in support of the nanoparticlequality, Tsai et al. [184] reported the photogeneration of E′γ centers reach-ing densities ≥ 1017 g−1 in high purity fused silica subjected to intense (40mJcm−2 pulses; ArF laser) 6.4 eV photon irradiation. The E′ productionefficiency was found to increase with higher OH content, the E′ density,however, staying well below [OH].

A last but perhaps highly pertinent observation is that the observedshift in gc over time as the sample is left in room ambient is accompaniedby a striking reduction in the E′ signal intensity. In the adapted state,the signal has reduced, in average, by factor of ∼2±0.5. The experimentalobserved decay rate shows a slight tendency to decrease with increasingTan, the variations, however, being well within experimental error.

Information on the hf interaction is well known to be a key point inatomic identification of point defects. Several attempts were made to re-solve some associated (29Si) hf structure, however, so far without success,which is attributed to the low density of occurring defects and strong signaloverlap.


12150 12160 12170 12180 12190 12200

(b)

(a)

1005 °C SiO-vac. anneal33.9 GHz; 100 K

E'γEX

g=1.99891

Si:P

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

Figure 2.11: Q-band ESR spectra of the fumed silica nanoparticles subjected toSiO-vac. annealing at 1005 ◦C measured at 100 K (a) immediate upon additionalVUV irradiation and (b) several days later, the sample being left in room ambient.The dashed curves represent spectra simulations employing Gaussian broadeningfunctions. The g values used for both simulations are (a) g1=2.00168, g2=2.00058,g3=2.0003; (b) g1=2.00171, g2=2.00053, g3=2.0003. The signal at g=1.99891stems from a comounted Si:P marker.

SiO impact

After the application of the SiO-vac. anneal at various temperatures fol-lowed by VUV irradiation, a characteristic central-line two-peak powderpattern signal is observed much similar (e.g., regarding width, overall shape,and g matrix) to that of the common E′γ variant [180, 46]. Fig. 2.11 showsrepresentative Q-band spectra recorded on a Tan∼1005 ◦C sample (a) im-mediately upon VUV activation and (b) several days later, the sample beingleft in room ambient. Both spectra could be well simulated (cf. Fig. 2.11,dashed curves) using similar ESR parameters (g value and constituent linewidths).

For the various Tan an average drop in E′ density by a factor ∼1.6±0.5is observed after leaving the sample for substantial time in room ambient.Similar to the case of fumed silica subjected to simple vacuum annealing,the experimental observed decay rate shows a slight tendency to decrease


0.85 0.90 0.95 1.00

1E15

1E16

1E17

1000105011001150

Tan (°C)E'

γ def

ect d

ensi

ty (c

m-3)

1000/Tan (1/°C)

1200

Figure 2.12: Defect densities of the E′ defect inferred from K and Q-band ESRin fumed silica subjected to a ∼3 h SiO-vac. anneal (solid red symbols) or a ∼1 hstandard vacuum anneal (open black symbols) at various temperatures subjected to∼1 h VUV irradiation. The measurements were performed immediately after theVUV activation. The dashed curve is merely meant to guide the eye.

with increasing Tan, the variations being well within experimental error.A main finding here is that compared to ordinary vacuum annealing, theSiO-vac. anneal results in a drastic enhancement of E′ density, increasingwith Tan to a maximum defect density of (4±1)×1017 cm−3 attained forTan=1205 ◦C, as illustrated by Fig. 2.12. For the sample SiO-vac. annealedat Tan∼1105 ◦C the attained E′ density is over a 100 times larger than forthe standard annealed sample.

Another interesting observation appears from Fig. 2.13 showing zoomedin representative Q-band ESR spectra of fumed silica nanoparticles sub-jected to SiO-vac. annealing [Fig. 2.13 (a)] and conventional vacuum an-nealing [Fig. 2.13 (b)] at Tan∼1005 ◦C recorded immediate upon VUVirradiation. From this comparison it can be seen that the E′ gc of thenanoparticles subjected to the SiO-vac. anneal has not noticeably shiftedwith respect to the general bulk value (gc∼2.00058), in contrast with thegc for the sample subjected to the vacuum anneal that has shifted upward


12170 12175 12180 12185 12190 12195 12200

2.00065

2.00056

g=1.99891

Si:PE'

γ

(b)

(a)

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

1005 °C anneal + VUV33.9 GHz: 100 K

Figure 2.13: Powder pattern Q-band ESR spectra of the E′ signal measured uponVUV activation of the fumed silica nanoparticles subjected to (a) a SiO-vac. annealat 1005 ◦C and (b) a standard vacuum anneal at 1005 ◦C. The dashed curvesrepresent spectra simulations. The lower S/N ratio for spectrum (a) compared to(b) is due to the smaller amount of sample available.

to ∼2.00065.

Hyperfine structure

As mentioned before, caused by the low E′ density encountered even afterexhaustive VUV irradiation of the as-received and vacuum annealed fumedsilica samples, we were unable to discern any hf splitting originating fromthe weakly abundant 29Si (4.7 at.%) nuclei. The realized increase in E′

density upon SiO-vac. annealing with additional VUV irradiation, however,enabled us to observe the 29Si hf structure related to E′ defects. Figure 2.14(a) shows a typical high-power second harmonic mode Q-band spectrumobserved for the fumed silica nanoparticles subjected to SiO-vac. annealingat 1105 ◦C and additional VUV irradiation after leaving the sample for daysin room ambient. For reference we also performed a similar measurement[Fig. 2.14 (b)] on suprasil I synthetic silica subjected to 100 Mrad(Si)60Co γ-irradiation at 300 K. In both samples two doublet structures areobserved. The central line, largely off scale, corresponds to the E′γ centers


11900 12000 12100 12200 12300 12400 12500

~ 73 G

~ 418 G

~ 438 G

34.1 GHz; 100 K

dP

µ/dB

(arb

. uni

ts)

magnetic field (G)

(b)

(a)

Figure 2.14: High power (saturation) second harmonic Q-band spectra for (a)fumed silica subjected to a ∼1105 ◦C SiO-vac. anneal and additional VUVirradiation measured at 100 K after leaving the samples for substantial timein room ambient and (b) Suprasil I synthetic silica subjected to 100 Mrad(Si)60Co γ-irradiation. The red dashed curves represent simulations of the primary29Si hf structure of the E′ centers, obtained using the principal g matrix valuesg1=2.0017, g2=2.0005, g3=2.0002 and hf parameters (a) Aiso=436 G, b=22 Gand (b) Aiso=418 G, b=22 G. Spectra were aligned at the resonance field of theE′ central resonance.


with no hf splitting (unpaired electron located on 28Si or 30Si nuclei ofI=0). The broad absorption line overlapping with the low-field parts ofthe doublets originates from OHCs. The inner hf doublet –henceforthreferred to as the 73-G doublet– exhibits a splitting of 69±4 G and 73±2 Gfor the nanoparticles and suprasil, respectively. More zoomed-in spectra,better displaying the inner doublet in fumed silica and suprasil are shownin Fig. 2.15. As elaborated on in section 1.2.3, a similar doublet has beenobserved previously in various types of a-SiO2 and was attributed [15, 84]to O2=Si•-H entities with an H atom in a back bond instead of O. Itshould be noted that this doublet appears less prominent in the fumedsilica powder than in the suprasil sample. The outer doublet (referred toas 418-G doublet) is, within the lines of this work, of more interest as itconcerns the well-known primary 29Si hf splitting of the E′ center [51, 47].The measured hf splittings are 438±2 G and 418±2 G for the nanoparticles

12075 12100 12125 12150 12175 12200 12225 12250

(b)

(a)

34.1 GHz; 100 K

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

A=68 G

A=73 G

Figure 2.15: Zoomed-in high power (saturation) second harmonic Q-band spectrafor (a) fumed silica subjected to a ∼1105 ◦C SiO-vac. anneal and additionalVUV irradiation measured at 100 K after leaving the samples for substantial timein room ambient and (b) Suprasil I synthetic silica subjected to 100 Mrad(Si) 60Coγ-irradiation. The blue dashed curves represent simulations of the 1H hf structureof the H associated E′ centers (O2=Si•–H), obtained using the isotropic g valueg=2.0019 and hf parameter (a) A=69 G and (b) A=73 G. Spectra were aligned atthe resonance field of the E′ central resonance, and scaled to equal E′ intensity.


and suprasil, respectively (cf. Fig. 2.14), bearing out a noticeable increasein hf spitting for fumed silica.

g-value considerations

The adapted state

For purposes of assessing possible specific structural characteristics pertain-ing to the a-SiO2 network of the studied nm-sized silica particles, it mayappear interesting to analyze what the revealed distinct attributes of theobserved E′ signal, such as the principal g matrix values, may tell us aboutthe centers’ local environment within the a-SiO2 matrix they are imbeddedin. But before starting, as several distinct species of E′ centers have beendelineated over the years in glassy SiO2, a basic quest concerns what partic-ular type(s) of E′ center we are actually dealing with in these fumed silicaparticles. Here, key information concerns the ESR observations on the par-ticles in the adapted state or after post-manufacture heating at Tan≥1100◦C, exposing the central E′ Zeeman resonances to converge to a commonsignal, and there is little doubt that this concerns the familiar E′γ variant:This conclusion is based on the observed g values (g matrix, gc) and the ex-hibited (nearly axially symmetric) powder pattern line shape as supportedby detailed computer simulations of the observed signals [180, 46, 47].

More evidence for this conclusion is adduced by Table 1.2 where rep-resentative g value data are assembled for the relevant most prominentE′γ-type centers reported in the literature and Table 2.2 were the currentE′γ (E′nano) results are presented. From this comparison also, it followsthat the SiO2 nanoparticles E′ signal does not concern the axially sym-metric E′β (Refs [48]) and E′s (Refs. [49, 185]) variant nor the extreme(regarding g matrix) orthorhombic E′α type [48]. One may remark, how-ever, that the comparison of g matrix elements usually reported in theliterature situates in the last significant digit, i.e., 10−4 (See, e.g., Ref. [4]).Generally, without having taken special measures (e.g., usage of accurateabsolute g marker samples or ingenious high accuracy methods for in situmeasurement of the at-the-sample magnetic field), most g value specifica-tions do not extend beyond the fourth decimal digit (corresponding to anabsolute ~B field determination accuracy of ∼0.16 G and ∼0.36 G for Xand K-band, respectively). So, over the many g data in the literature, oneencounters (slight) variations in the values reported for identical defects,urging some caution when intending to carry out comparisons –in particu-lar, when looking for small systematic variations. To settle this, the only


Table 2.2: Comparison of experimental principal axis g matrix values of genericE′ centers observed in this work. The g values are determined relative to a Si:P g

marker. The accuracy on g is ±0.00004.

E′γ [a] E′nano(bulk) [b] E′nano(surface) [c]g1 2.00175 2.00174 2.00174g2 2.00056 2.00048 2.00070g3 2.00030 2.00020 2.00030

[a] Observed in suprasil I synthetic silica –obtained from Heraeus (Germany) con-taining [OH]∼1200 ppm (by weight), [H20]≤600 ppm, and total metal content <1ppm– subjected to 100 Mrad(Si) 60Co γ-rays at 300 K.[b] Observed in fumed silica 7 nm particles subjected to 10 eV photons obtainedfrom a Kr-resonant discharge lamp (flux∼1015 cm−2s−1) after leaving the samplefor substantial time in room ambient, i.e., in the adapted state.[c] Observed in fumed silica 7 nm particles immediate upon 10 eV VUV, pho-tons obtained from a Kr-resonant discharge lamp (flux∼1015 cm−2s−1), i.e., inthe non-adapted state.

way out appears to first establish internal consistency. Thus, for normal-ization (calibration) purposes we have additionally calibrated within ourESR approach based on the usage of a single high accuracy g marker, the’standard’ most widely studied E′ center produced in synthetic fused silicasuprasil I by 100 Mrad(Si) 60Co γ-irradiation. The recorded K-band spec-trum is shown in Fig. 2.8. The g matrix element results, consistent withinan absolute accuracy of 2×10−6 between K and Q-band experiments, areincluded in Table 2.2 as well. Finally, in terms of g data, Table 2.2 alsoputs in better perspective the difference between the observed E′ systemsin fumed SiO2 particles in the adapted and non-adapted state.

The non-adapted state

As the key point of focus, what then causes the notable variations in E′

features –that is, shift in gc with attendant variation in powder patternshape– for the non-adapted (other) sample states, i.e., in the as-receivedstate or after heating in the Tan.1100 ◦C range, both just after VUVirradiation? Obviously, there may be various reasons.

One pertinent experimental observation is that the shift in gc during postVUV excitation decay or after heating at Tan>850 ◦C appears persistently


accompanied by a decrease in overall E′ intensity, which would indicatethat part of the ESR active centers drop out. The further interpretationcould then take two points of view: A first one could start from the occur-rence of only one E′γ system, yet characterized by a profound distribution inESR broadening parameters. The applied VUV excitation would efficientlyESR activate centers all over the distribution. Then, during the subsequentstay in room ambient (adaptation period), part of the E′ centers, perhapsdue to environmental influence, would become gradually ESR inactivated(passivated), predominantly starting with those centers pertaining to the(extreme) tails of the ESR parameter distributions. Thus, with progressinginactivation, the more standard E′γ signal would naturally surface, charac-teristic of macroscopic glassy SiO2 with more standard spreads in relevantESR broadening parameters.

In a second scenario, the more distorted part of the E′γ centers, mak-ing up the extended tails of the g distributions, could be rather seen asstemming from a separate, second E′ system. In an idealized scheme, onecould picture the first, more bulk a-SiO2-like E′γ system as constituting thecenters located more in the center (”bulk”) of the nm-sized silica particles;The second system would comprise the defects pertaining to the structurallymore distorted (strained) surface and near surface layers –the surface E′

system– more apt to environmental chemical and physical interaction (ad-sorption) impact. (About 23% of the molecules belongs to a monolayer(∼0.29 nm) of SiO2 entities at the surface of a 7-nm SiO2 particle, as vi-sualized in Fig. 2.16.) Thus, the E′ signal might then rather be consideredas the superposition of (at least) two overlapping signals originating fromtwo separate E′ defect systems. The first bath is dominant in the spectraof the particles in the adapted state or after post-manufacture heating atTan≥1100 ◦C leading to the observation of signals with very similar prin-cipal g values and g value distributions as the E′ systems observed in bulksilica. The surface E′ system is dominant in the spectra of the as-receivedsample immediate after VUV irradiation giving rise to a shift in gc and anoverall more broadened spectrum.

For the remainder, the decay (adaptation) scenario upon VUV activa-tion would be similar to the previous view: The second surface E′ systemwould be gradually inactivated with increasing dwell time in room ambient,ultimately leaving only the first E′γ bath. (The time delay may result fromthe altering accessibility by ambient species of E′ centers located deeperin near surface layers). Interestingly, such a decrease in E′ signal intensityhas been observed before in the study of quartz crushed in vacuum [49].


~7 nm

~0.29 nm CORE

Figure 2.16: Schematic visualization of the surface area of a ∼7 nm fumed silicaparticle.

There, an E′γ-like (but not identical) signal was observed which could bereadily modified by interaction with active gases, such as CO2 and air.It is taken as evidence for the occurrence of surface dangling Si orbitals,i.e., the surface E′ center [46, 183, 49, 185], termed E′s, assigned to thehemicenter structure (O3≡Si•) at the quartz grain surface. A similar sen-sitivity of Si dangling hybrids (Si3≡Si•; D center) to the ambient has longago been demonstrated at the surface of Si in cracks penetrating from thesurface [186].

The surface E′ systems observed in crushed silica particles and thin de-posited a-SiO2 films on Si are characterized by a closely E′γ-like g matrix,yet with a slightly smaller gc value (∼2.0003) [46, 49] than the gc value gen-erally observed for the E′γ center in bulk silica’s (gc∼2.00058), which wouldconflict with the current observations (gc∼2.0076). However, here it shouldbe remarked that in spectroscopic terms, as outlined above, the measuredupward shift in gc may well result from a substantial distribution in ESRparameters. Such spreads are well known to arise from amorphous-statedisorder [4, 187], reflecting structural and strain induced site-to-site vari-ations in Si–O–Si bond angles. Besides, it should be added that currentlywe deal with surfaces of nm-sized particles, which may exhibit a distinctlydifferent dimension of distortion as compared to bulk-like grains and films.Thus, rather than counter, the current first report of such E′ system fromthe outer layers of nm-sized silica particles might just instead be taken asevidence exposing this.

There is, however, one E′-type surface center reported that does exhibit


a larger gc value than the gc value generally observed for the E′γ center inbulk silica (gc∼2.00058): Radzig et al. observed an E′-type defect centerwith principal g-values: g1=2.0003, g2=2.0007, and g3=2.0019 [188, 189].These values are very similar to those currently observed for the E′ centerslocated in the surface layers of fumed silica. Radzig attributed the defectto a O2=Si•–OH entity with an OH group in a back bond, located at thesurface of high purity quartz. Identification of the defect was accomplishedthrough the observation of the primary 29Si hf splitting (Aiso=413 G; b=19G) and the small 1H hf splitting (Aiso=2.2 G; b=0.6 G) [189]. The presenceof such a defect site at the surface of the fumed silica nanoparticles maynot be totally unexpected as extensive NMR studies [165] have pointedout that large concentrations of silanols (Si–OH groups) are present atthe surface of fumed silica. If the observed surface E′-type defects in thenanoparticles would have such a OH group in the back bond, the central lineshould show additional structure resulting from the hf interaction with the∼100% abundant 1H nucleus. Yet, no such features could be observed inthe K and Q-band spectra recorded immediate upon VUV activation of theas-received samples. However, it is possible that when measuring at these(high) frequencies the hf splitting remains unresolved as a result of intrinsicline broadening. To overcome this difficulty, X-band measurements wereperformed resulting in a narrow powder pattern, as shown in Fig. 2.7. Ifpresent, X-band observation should provide the best chance to resolve suchhf splitting resulting from the H nucleus [190]. Yet, no such features areobserved from where we conclude that for the fumed silica nanoparticlesthe observed gc shift for the E′ centers located in the surface layers doesnot concern a back bonded OH group.

The overall broadened spectra observed for the as-received sample stateimmediately after VUV irradiation and the strong dependence on the an-neal temperature of the line width may also be understood as the resultof the presence of a surface E′ system. It seems reasonable to accept thatthe outer layers of the nanoparticles have a somewhat more distorted, morestrained structure as compared to the inner core of the particles. MD com-puter simulations did point out that the surface has a different density andstructure as compared to the interior of the cluster [156]. Furthermore, thesurface can be influenced (distorted) by the presence of adsorbed atomsor impurities, such as silanols [165]. The combined effect can lead to theobserved common broadening of the ESR spectra for the as-received statesample measured immediate upon VUV irradiation.

The observed dependence on observational temperature of the overall


line width can be explained by a temperature dependent surface strainadaptation. The thermal expansion coefficient αSiO2 of quartz, althoughoverall extremely low, does exhibit a clear temperature dependence, i.e., amonotonic increase, the largest relative variation (a factor ≥2000) occurringin the range [191] 6-75 K. Measuring at a higher temperature (77 K) canlead to some relaxation of the surface, thus decreasing the line width. Asecond possible explanation can be found with the adsorbed atoms. As thetemperature increases the atoms can start tumbling leading to motionalnarrowing [192]. As will be demonstrated later, tumbling methyl radicalsare indeed observed at these silica nanoparticles.

The effect of the post-manufacture heating on the E′ system may alsobe understood within the above outlined picture. In a first view, the re-sults may be explained by posing that with heating at increasingly highertemperatures in the range Tan≥850 ◦C, the surface E′ system is graduallyeliminated, through irreversible structural network adaptation or chemicalinteraction with impurities leading to firm bonding. The former may ap-pear less obvious, while the latter needs interference of impurities otherthan hydrogen. However, in a second more likely view, the surface E′

system might just plainly be irreversibly eliminated by heating-dependentreduction of the effective sample surface area through particle sintering.Such sintering is known to start from about 850 ◦C onward, as recentlyre-evidenced by Uchino et al. [153] exposing a decrease in specific surfacearea with Tan, well mimicking (with respect to the pertinent Tan range)the E′ gc-vs-Tan dependence currently measured immediately upon VUVactivation. This can be visualized by comparing Figs. 2.1 and 2.9. Whilewe cannot discriminate between the two interpretations, the latter, giventhe physical facts, must at least account for part of the E′ observations.

All in all, as a main result, the current data, concerning the central Zee-man resonance of the E′ centers, reveal structural aspects of the nm-sizedfumed silica particles different from bulk glassy SiO2. For one, g matrix al-terations are observed, assigned to enhanced amorphous state disorder andstrain: Structural variations affect the spatial distribution of the unpairedelectron orbital, which in turn, through the spin-orbit coupling, reflectsback into the observed g factor. It is interesting to note that in recentwork [193], small relative variations in g matrix elements (principal g val-ues) observed in dry silica as a function of increasing 60Co γ-irradiation(10−1-104 kGy) have also been attributed to structural modifications re-flecting back in the unpaired electron orbital, thus resulting in g shiftsthrough the spin-orbit coupling.


In a more direct structural picture, it is known that for tetragonallyarranged defects, such as the Si3≡Si• defect (Pb center) at the Si/SiO2

interface, shifts in, e.g., g‖ away from the average value may be directlyrelated to variations in the unpaired bond angle θ (angle between the un-paired sp3-like orbital direction and a Si–Si back bond). An increase in θresults in a shift in zero crossing g value towards higher values [194, 195].Shifts in g can thus be directly translated into a more planar or pyramidaldefect configuration [173, 185], which may provide useful information, alsoin the case of the E′ center, e.g., in terms of local stress. However, no suchclear relationship appears as yet established for the E′ center. This willconcern a next step in research, implying experimental refinement as wellas, perhaps more crucially, substantial theoretical effort.

SiO impact

To analyze the influence of the presence of SiO during the SiO-vac. annealon the fumed silica network, we compared our results obtained for theSiO-vac. anneal with those obtained for the regular vacuum anneals inthe range Tan∼1005-1205 ◦C. From Fig. 2.12 it can be seen that withthe presence of SiO during the anneal, the E′ density in the fumed silicaparticles has increased. This increase becomes gradually more drastic athigher anneal temperatures. Within the interpretation of interaction ofthe fumed silica network with released SiO molecules, yet without enteringinto atomic structural detail, this result might be understood on grounds ofsimple general stoichiometry: The incorporation of SiO in a SiO2(x) networkwill result in a net Si enrichment (or, equivalently, O deficiency). Thus, inthe fumed silica nanoparticles, which have been evidenced to be O rich(low E′ defect density), the incorporation of SiO might thus result in anincrease of the number of oxygen vacancies, hence E′ defect centers. ForTan∼1205 ◦C the total E′ density becomes comparable to, even slightlyhigher than, the density obtained for bulk thermal SiO2 (∼1.4×1017 g−1;standard quality) [181, 182]. It appears that the incorporation of SiO inthe fumed silica network causes a serious degradation of the oxide qualityin terms of oxygen vacancies. A similar increase in E′ density has beenobserved before ascribed to the interaction of SiO with the SiO2 networkat elevated temperatures [63, 65]. However, in those cases the ESR spectraobtained for Tan≥1005 ◦C were dominated by the S-center suggested tobe of the type SinO3−n≡Si• (n=1, 2) [63, 66]. No S center was observed inthe current fumed silica sample indicating that the nanoparticle networkstructure is unfavorable for S center creation.


Besides the change in defect density ascribed to SiO impact, anothernoteworthy difference concerns the observed gc value of the E′ signal, asillustrated by Fig. 2.13. For the standard vacuum anneal the gc value hadshifted to a somewhat higher value than the one typically observed in ther-mally grown SiO2 on Si (∆gc∼0.0001) due to the presence of E′ centers inthe surface regions of the nanoparticles, as discussed in the previous sec-tion. In the case where the SiO-vac. anneal is applied, no such shift couldbe observed, indicating that there is no significant fraction of surface E′

centers left after the SiO-vac. anneal, in which case more of the E′ centersin the (near) surface region would have turned into bulk silica properties bythe SiO impact. In another opinion this could indicate that while a largerfraction of surface E′ centers would still be left, their contribution to theoverall observed E′ signal is insufficient to cause a significant shift in theobserved gc. This would mean that a fraction of ∼40% is insufficient toexperimentally induce a measurable g shift, as indeed confirmed by inde-pendent spectral simulations. This may find some support in the generallyexperimentally observed trend of a decreasing decay rate with increasingTan. Anyhow, it can be stated that the E′ centers generated during theSiO-vac. anneal exhibit predominantly core region properties. The surfaceregions and the core region appear to respond differently to penetratingSiO molecules.

The primary 29Si hyperfine interaction of the E′γ-centers

Observed structure

In the previous section, it has been pointed out that monitoring the prop-erties (e.g., gc value) of the central Zeeman ESR signal of embedded intrin-sic E′ centers as a function of post-manufacture annealing and treatment(aging, VUV, and UV excitation) may provide useful information on thenetwork structure of silica nanoparticles. As to the core region of these,no variation in the E′ Zeeman signal, i.e., gc shift, was observed comparedto bulk a-SiO2, indicating that the core is quite bulk silica-like. Uchinoet al., however, obtained experimental infrared and Raman data [153] in-dicating that the structure of the core of the nanoparticles is not fullyidentical to that of bulk silica. If so, most probably then these alterationsin the network are too small to engender a detectable gc shift of embeddedE′ centers, leaving this ’ESR route’ unsuccessful. Here, the currently ob-served hf structure may come to the rescue (cf. Fig. 2.14). The signals areobserved on fumed nanoparticle samples subjected to SiO-vac. annealing


in room ambient for substantial time, the latter resulting in suppressionof the presumed surface E′ system. So, the observed E′ hf structure (twodoublets) would also pertain to the core part of the nanoparticles.

(a)Inner doublet : We can be short about the inner 73-G doublet, ac-tually observed with splitting of 69±4 G and 73±2 G in the nanoparticlesand suprasil reference sample, respectively. Extracting the hf parameters ofthis doublet from the fumed silica spectrum is strongly hampered throughinterference of the co-present OHC signal with the low-field doublet sig-nal, as can be seen from Fig. 2.14, resulting in a reduced accuracy (±4 Gon the inferred hf splitting). To get more accurate information on the hfsplitting spectral simulations were carried out using a code based on exactmatrix diagonalization incorporating the Breit-Rabi formula. For this end,the high-power second harmonic spectra were assumed to resemble directabsorption shapes, an observation commented on previously [14], for whichempirical evidence has been provided. For suprasil the observed hf doubletcould be well simulated, as shown in Fig. 2.15, using the isotropic param-eters g=2.0019 and Aiso=73 G; any anisotropic contributions remainedunresolved. As to fumed silica, independent fitting of the hf doublet waseven more severely obstructed by the overlapping OHC signal. However,by assuming that the central g value of the hydrogen associated E′ centeris the same in suprasil and fumed silica (as it is for E′γ), we may circum-vent this hurdle. Using this in the fitting procedure and focusing on thehigh-field doublet signal, we obtained Aiso=68 G with an estimated accu-racy of ±2 G. In principle, like the primary 418-G doublet, the propertiesof this 73-G doublet could be perused as well to infer information aboutthe properties of the core of the nanoparticles where the originating defects(O2=Si•–H) are built in (vide infra). Indeed, although on the limit of ex-perimental accuracy, there may be a decrease in hf splitting compared tothe reference sample. Yet the doublet seems less amenable for that purposefor reasons such as attainable experimental accuracy (signal overlap) andthe absence of a reference base linking hf structure parameters with SiO2

network/material properties. Hence, we refrain from further hf analysis.There is, however, one more interesting observation concerning the rel-

ative intensities of the H-doublet in fumed silica and suprasil. Observedmany times before in various types of a-SiO2, this 73-G doublet was as-signed to the H-associated E′-type center O2=Si•–H, the hf splitting thusoriginating from the interaction of the unpaired electron with a proton ina back bond [15, 84]. The defect thus signals the presence (though notexclusively, of course) of hydrogen in the silica network. In the second har-


monic mode Q-band spectra (cf. Figs. 2.14 and 2.15), this doublet appearsrelatively more prominent for suprasil than for fumed silica. For equalspectroscopic behavior, this suggests that fumed silica nanoparticles wouldcontain less H. It should be added that this conclusion also assumes thatthe ESR-activation ratio of E′γ/O2=Si•–H defects is effectuated equally byVUV and γ-irradiation (applied to fumed silica and suprasil, respectively).Though perhaps plausible, this does not have to be a priori so, and somecaution may be needed.

(b) Outer doublet : The observed second outer doublet concerns the pri-mary 29Si hf splitting of E′ centers [51, 47] located in the core of thenanoparticles. The hf parameters are analyzed to attain more in depthinformation on the core structure. From Fig. 2.14 it can clearly be seenthat the primary 29Si hf splitting of the E′ center is larger in the fumedsilica nanoparticles than for suprasil. To obtain more detailed informationabout the hf structure spectral simulations were carried out. For the bulksuprasil sample the observed hf doublet could be well simulated using simi-lar 29Si hf parameters (A‖=462 G and A⊥=396 G) as reported by Griscomet al. for glassy silica [51]. The g values used for this fitting procedurewere taken from the simulation of the central line: g1=2.0017, g2=2.0005,and g3=2.0002. The parallel (A‖) and perpendicular (A⊥) hf constants arelinked to the isotropic (Aiso) part (s part: Fermi contact interaction) andthe anisotropic (b) part (p part: dipolar interaction) of the hf interactionthrough the equations:

A‖ = Aiso + 2b, (2.2)A⊥ = Aiso − b.

For the suprasil E′ center this results in Aiso=418 G and b=22 G.As can be seen from Fig. 2.14, the co-present OHC signal is badly

interfering with the low-field doublet signal, particularly so for the fumednanoparticles, exhibiting a relatively stronger OHC component. This ham-pered independent fitting of the primary 29Si hf doublet for fumed silica.Therefore, based on various hf constants reported in the literature [51, 196],and in a similar attitude as taken before we assumed that the increase inthe measured hf splitting for the E′ center in fumed silica is mainly causedby an increase in the isotropic part of the hf interaction. Thus, in the fittingprocedure b (=22 G) was kept constant, which enables a clear comparisonwith data from the literature. For the primary 29Si hf parameters of the E′

center in the fumed silica nanoparticles, this led to the results: A‖=480 Gand A⊥=414 G or, put differently, Aiso=436 G and b=22 G. The inferred


Aiso and b values provide access to the structure of the corresponding E′

centers.

E′ defect structure

The first researchers to observe the primary hyperfine of the E′γ centers ina-SiO2 were Griscom et al. [51]. From their results, the authors obtaineddetailed information about the short-range structure of the amorphous net-work, such as, e.g., the O–Si–defect bond angle θ, using the simple expres-sion

tan(θ) = −[2(1 +b

Ap

AsAiso

)]12 , (2.3)

where As and Ap are the hf coupling constants for 3s and 3p electrons onneutral atomic silicon. Using the values for As and Ap cited by Mortonand Preston [8], as elaborated on in section 1.1.1, the values obtained forthe bond angle θ in the current work are θbulk=111.86◦ and θnano=112.15◦

for the E′ centers in suprasil (θbulk) and fumed silica (θnano), respectively.Assuming an accuracy of ±2 G on the experimentally obtained isotropichf constants, the accuracy obtained on θ is ±0.03◦. The inferred valuefor θ is somewhat larger (∆θ = θnano − θbulk) for the E′ centers in thecore of the nanoparticles pointing to a configurationally more pyramidal E′

defect structure, as illustrated in Fig. 2.17. As to this conclusion, one mayconjecture that observation of the 29Si E′ hf structure has been enabledthrough interaction with SiO at high Tan (E′ density enhancement), whichmay have resulted in significant modification of the SiO2 network so thatthe current results would not pertain to the pristine nanoparticle properties.The impact, however, on the global network structure is considered minor

θbulk θnano>θbulk

Si

O

(a) (b)

Figure 2.17: Visualization of the structural modification of the E′ center genericunit in fumed silica (b) nanoparticles to a more pyramidal defect structure com-pared to bulk silica (a).


given the still low maximum density of E′ defects (∼1×1017 cm−3) in theSiO2 matrix (∼5 at. ppm) of the fume samples for which the hf data arecurrently reported (Tan∼1105 ◦C).

There is still one more pertinent remark that should be made concerningthis defect bond angle analysis. Equation 2.3 is based on a simplified theorybased on various assumptions including perfect sp3 hybridization. It shouldbe noted that in literature different bond angle–hf relationships exist (start-ing from different assumptions) resulting in different values for θ. Edwardsand Fowler [197] for instance argued that in a relative ionic system such asSiO2, the use of neutral atomic wave functions in Eq. 2.3 is inappropriate.They suggested the use of the wave functions from Si(+1), as a way to in-clude the ionicity of the sp3 hybridization in SiO2. The ensuing modifiedθ–hf parameter relation results, as verified in our case, in (slightly) highervalues for θ (θnano=113.4◦ and θbulk=113.7◦ for the E′ center in fumed sil-ica and suprasil, respectively) but still similar values for ∆θ are obtained.Pacchioni and Vitiello [198] obtained from their first-principle Hartree-Fockand DFT calculations a much stronger relationship between Aiso and θ. Forthe range of values from 100◦ to 110◦, they obtained an increase in Aisoof 6 G for a decrease in θ of 1◦. It appears that the Aiso–θ relationship isvery responsive to various structural/morphological parameters. However,the trends obtained even from the simplest theoretical analysis are at leastqualitatively in agreement: an increase in Aiso is a result of an increase inθ. On this notion it may still be appropriate to continue using the resultsobtained from Eq. 2.3.

Having attained a specific quantification for a change in the structureof the E′ center between bulk silica and the core of silica nanoparticles,we may want to transfer this information to assessing how it would affectother experimental quantities, such as, the g value of the ESR resonance.As already mentioned, generally, such a structural modification does notonly affect the hf splitting but will also result in a g shift. However, forthe core E′ centers no shift could be detected, which then must mean thatthe variations in θ remain most probably too small to result in a detectableg shift. For the surface E′ system a g shift (increase) was observed, indi-cating that the variations for θ would be stronger for the outer layers ofthe nanoparticles. But here, whether the surface E′ centers have a moreplanar or pyramidal defect structure remains unknown as no hf structurecould be resolved due to sensitivity reasons. So, unfortunately, no directexperimental link can be provided between a shift in g and a change instructural pyramidality. The answer may be provided by theory. However,


if assuming that the E′ centers in the surface layers follow the same trend,but more pronounced, as their counterparts located in the core regions,i.e., enhanced pyramidality, than we would reach the conclusion that anincrease in θ is accompanied by an increase in g. This result would beconsistent with theoretical work reported for the unpaired Si bonds in a-Siand the Pb center [194, 195], where an increase in gc value is linked withan increase in θ and hence a more pyramidal defect structure.

On the other hand, for the surface E′ system (E′s) observed in crushedsilica particles discussed before [49], a smaller gc∼2.0003 was observed ac-companied by an increase in Aiso to ∼466 G. If the surface E′ system infumed silica behaves in a similar way, the observed increase in gc wouldbe indicative for a decrease in Aiso and consequently in θ and thus a moreplanar E′ defect structure. Within this picture the surface region and coreregion of the nanoparticles would exhibit an opposite structural behavior,e.g., where the core region is more strained, densified (vide infra), the sur-face layers would be more relaxed compared to bulk SiO2.

Densification

While the attained atomic scale information on the structure of a spe-cific point defect is definitely of interest on its own right, the next stepbeyond in analysis reaches a main goal: Very basically, it touches the po-tentiality of (intrinsic) point defects in providing pertinent information onlarger scale structural/material physical quantities of the solid matrix theseare embedded in –what can an atomic scale point defect tell us about itsbroader environment? The potentiality of point defects in this sense isdemonstrated by many examples before, e.g., Pb defects at Si/SiO2 inter-faces [29, 202, 173], Fe-impurities in Si [203]. Assessing this is a main goal.As to the current nanoparticles, the increase observed in Aiso for the E′

centers residing in the core of the nanoparticles as compared to bulk silicamay also provide in depth information on more global structural/materialparameters of the matrix the defects are embedded in. As a starting pointwe have compiled in Table 2.3 the results of the primary 29Si hf splitting ofthe E′ center for the current nanoparticles and for glassy SiO2 reported inthe literature [172, 51, 196, 199, 14, 15, 200, 201]. Over the various values,excluding densified and 29Si implanted samples, an average hf splitting of419±4 G is attained. A substantial increase in splitting is seen to resultfrom densification, as well as, though somewhat less, from Si implantation.As the latter was argued also to effectuate densification [201, 200], there ap-pears a clear link between densification and enhancement of 29Si hf splitting.


Table 2.3: Comparison of the experimental primary 29Si hyperfine splitting of E′γcenters in silica glass.

Hyperfine Ref. Sample Damaging agentsplitting (G)

423 [51] Corning 7940 [a]423 [51] Corning 7943 [a]423 [51, 14] Suprasil I [a]423 [51] Suprasil N-I [a]423 [51] Dynasil 1000 [a]423 [51] NRL high purity [a]

precipitated silica418 [51] 95% 29Si enriched SiO2 [a]420 [14] 95.3% 29Si enriched SiO2 [a]420 [14] 99.8% 29Si enriched SiO2 [a]418 [15] High OH silica spectrosil [a]

rods (∼1200 ppm OH)418 [15] Suprasil 2 powder; [a]

high OH + OD410 [196] Suprasil I [b]410 [196] Suprasil W1 [b]465 [196] Suprasil I; ∼24% densified [b]465 [196] Suprasil W1; ∼24% densified [b]465 [199] Suprasil I; ∼24% densified [c]440 [200] Wet synthetic SiO2 glass

(type II, [OH]∼3×1019 cm−3);29Si implanted

420 [172] Fused silica glass powder [d]420 [172] Suprasil I [d]440 [201] Dry a-SiO2 (type II,

[OH]≤1017 cm−3);29Si implanted

418±2 This work Suprasil I synthetic silica [a]438±2 This work Fumed 7 nm SiO2 particles; [e]

1105 ◦C SiO-vac. anneal

[a] ∼100 Mrad(Si) 60Co γ-rays.[b] ∼76 Mrad(Si) 60Co γ-rays.[c] 248-nm KrF laser irradiation.[d] ∼8.3×102 Mrad(Si) 60Co γ-rays.[e] 10 eV photons obtained from a Kr-resonant discharge lamp (flux∼1015

cm−2s−1).


Figure 2.18: Fractional increase of the experimentally obtained isotropic hf con-stant as a function of densification of the silica sample. The increase is referredto the value of 410 G found in undensified Suprasil. The cross refers to a SuprasilW1 sample and is included to demonstrate that the type of silica does not play arole. The figure is taken from Ref. [196].

Thus, such an increase in Aiso has been observed before [196, 199, 200, 201]and is assigned to a densification of the SiO2-matrix.

In a remarkable study of the hf splitting of the E′ center in a-SiO2

as a function of the density [196] for samples densified under hydrostaticpressure at high temperatures, Devine et al. reported an increase in Aisowith growing silica density, as illustrated in Fig. 2.18. Densification of thesilica of 24% was observed to result in an increase of the hf splitting to∼465 G, from where an increase was inferred in the length of the backbonds associated with the defected Si atom at the oxygen vacancy site,a reduction of the mean bridging Si–O–Si bond angle of the defect backbonds, and a decrease of the tetrahedral O–Si–O bond angle of the defect.These results did not depend on the type of silica studied, so it seemsreasonable that their inferences can also be applied for the analysis of thecurrent fumed silica data. In this view the obtained fractional increase inAiso of 4.1% would imply a densification of the fumed silica network of 5.9%,corresponding to an increase in specific gravity to 2.33 gcm−3 as comparedto 2.202 gcm−3 for undensified silica. The core of the nanoparticles arethus found to be densified.

But how does this densification come about in terms of modification ofthe SiO2 network? In addition to ESR, Devine et al. also performed x-ray


scattering, Raman and NMR measurements on densified silica [204]. Fromthese results they concluded that the densification produces a decrease inthe oxygen-second-nearest-neighbor-oxygen separation (R). So, R is theminimum distance of the closest approach of two oxygen atoms belonging toadjacent SiO4 tetrahedra. As a next crucial link, one calculation concludesthat restrictions on interoxygen distances impose limits on possibly occur-ring ring configurations of SiO4 tetrahedra in the SiO2 network [205]. Thedescription of the a-silica network in terms of the distribution of occurringn-membered rings (n=3-6) is a well known approach [See, e.g., Ref. [206]for an excellent overview], and there is a general consensus the structure tocontain ring structures of different sizes [206, 207]. The average ring sizein standard bulk vitreous SiO2 is ∼6 [206]. The calculations infer that adecrease in R is indicative for the presence of relatively more low memberedrings. At least the existence of such configurations is no longer excludedthrough distance considerations. On the basis of the previous inferences,our E′ hf ESR data indicate an enhanced density of lower membered (SiO4

tetrahedra) rings in the core part of silica nanoparticles compared to bulksilica. It should be borne in mind, however, that the inferences made byDevine et al. are based on various empirical relationships, such as, e.g.,between bond angles and bond lengths. As Edwards and Fowler [197] putit: The trends predicted by this analysis are almost certainly qualitativelycorrect, but the relationships used are inappropriate.

Generally, we can conclude from our ESR results that the densificationof the core of the fumed silica nanoparticles results in an increase in the29Si hf splitting of the E′ centers located in the core, tentatively linkedto a modification of the distribution in occurring ring structures in thefumed silica network. These results confirm independently those previouslyreported by Uchino et al. that the core of the nanoparticles differs frombulk silica [153] in that it would be comprised of more low (n=3 and 4)membered rings. It thus appears that through the observation of the hfstructure of embedded point defects, ESR represents a powerful tool toprovide us with detailed information about the network structure of thenanoparticles on atomic-scale.

2.5.2 A newly isolated ESR signal

Observations

Upon post-manufacture heating an unidentified signal (termed LU2) newlyappears in maximum densities (6.0±0.6)×1015 g−1 (∼2.5×10−3 defect/particle),


33.5 GHz; 10K

20.5 GHz; 4.2K2.00276 g = 1.99869

Si:P

magnetic field (G)

dPµ/d

B (a

rb. u

nits

)

20 G

Figure 2.19: Powder pattern K-band (Pµ∼0.3 nW ; Bm∼0.5 G) and Q-band(Pµ∼40 nW ; Bm∼0.3 G) ESR spectra of the observed signal, labeled LU2, insilica nanoparticles annealed at Tan=960 ◦C. The dashed curves represent spectrasimulations (see text for parameters).

without the additional VUV irradiation, that is required for the detectionof the E′ center. At K-band it appears as a single asymmetric signal ofgc=2.00276±0.00005 and peak-to-peak width ∆Bpp=4±0.2 G. Figure 2.19shows K and Q-band ESR spectra observed at low T , after annealing atTan=960 ◦C. The signals exhibit powder pattern properties, and bothcould be readily consistently simulated (symbols in Fig. 2.19) by one effec-tive spin S=1/2 center, of closely axial symmetry, with one set of g matrixprincipal values given as g‖=2.0041 and g⊥=2.0027; Only the microwavefrequency ν has been changed and the Lorentzian broadening function hasbeen increased with ν. Other observations include: In the range 850-1115◦C, the signal intensity is found to be maximal for Tan∼960 ◦C; Subjectingthe sample to VUV irradiation decreases the intensity of the center.

It is realized, however, that the attained signal-to-noise ratio is limited.So, conforming to scientific rigor, full allowance should be made for otherinterpretations. For one, rather than adhering to the powder pattern of


one type of defect, e.g., the overlap of two separate signals of differentg values, which structure would also broaden with increasing microwavefrequency. Through computer simulations, this route has been carefullyanalyzed. However, despite fitting over a broad range of parameter values(g, ∆Bpp, line shape, relative intensity), we failed to realize a match assatisfactory as readily obtained by the single powder pattern interpretation,the main delimiting obstacle appearing to be the observed overall signalsymmetry, e.g., the signal measured at Q-band at 150 K constitutes awell expressed ’two peak’ powder pattern, as shown in Fig. 2.20. This iswhy the powder pattern interpretation is favored. Of course, it cannotbe excluded that with still increasing the number of independent signals(3 or more), thus concomitantly unrealistically augmenting the numberof independent fitting parameters (4 per signal), a satisfactory consistentfit might ultimately be realized –but, obviously, at the cost of scientificcredibility.

12080 12090 12100 12110 12120 12130 12140

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

960 °C vac. anneal33.5 GHz; 150 K

Si:PLU2

Figure 2.20: Q-band (Pµ∼0.16 mW ; Bm∼0.6 G) ESR spectra of the observedLU2 signal in silica nanoparticles annealed at Tan=960 ◦C, displaying a well ex-pressed ”two peak” powder pattern. The dashed curve represents the LU2 spectrumsimulation (see text for parameters).


Discussion

Without doubt, a main key question concerns the origin of this signal. As iswell known from previous studies on atomic identification of point defects,almost exclusively based on ESR analysis, conclusive determination of thenature of the originating defect necessarily requires information on the hfinteraction. Identification of the defect is considered much relevant as itsappearance may be closely tied to a specific characteristic of the fumedsilica network, possibly different from the bulk glassy SiO2 state. Severalattempts were made to resolve associated hf structure from the background,yet so far without success, which makes that the discussion of the center’sorigin remains largely speculative.

Purely based on gc, one possible candidate could be the S center (gc∼2.0027), tentatively assigned [62, 63, 66] to the E′-like centers SinO3−n≡Si•

(n=1,2) –likely SiO2≡Si•, as elaborated on in section 1.2.2. That center wasalso revealed after high-Tan annealing in vacuum, prominently so in ther-mally grown SiO2 layers on Si. However, at first sight, general line widthand g matrix asymmetry considerations would disfavor this assignment.

In looking further for other models, given the measured zero crossingvalue gc=2.00276, we may also consider the possibility the signal to orig-inate from C-related defects. Indeed, many signals observed around thatg value (see, e.g., Ref. [208]) have been attributed to (identified as) C-related defects, e.g., as encountered in carbonaceous materials such as me-chanically damaged [209] C powder (∆Bpp∼5.5 G; g=2.0027±0.0002) anda-SiC films (∆Bpp∼6 G; g=2.003±0.001) [210], and as contaminants onion-implanted glass [211] and various heat-treated surfaces [212], the lat-ter, for quartz and glass, characterized by a width ∆Bpp(X-band)∼1 G andg=2.0028. A singlet characterized by an approximately Lorentzian shape,∆Bpp(X-band; 300 K)∼1-3 G and g=2.0026±0.0002 was reported in vir-tually all ion-implanted glasses [211] and ascribed to surface C contami-nation. Along the case of heat-treated surfaces, the observed LU2 signalcould have resulted from the annealing in vacuum. However, counter tothis interpretation would appear the currently applied high Tan values andthe non-disappearance of the ESR signal when exposed to air (room am-bient) [212]. But it is well realized that for unequal circumstances (e.g.,material, shape, treatment) signals may appear different, and a possiblerelationship of LU2 to carbon is not ruled out; However, it cannot be sub-stantiated any further either. Thus, conclusive identification of this (new)defect remains still open.


7300 7320 7340 7360 7380 7400 7420

2.0074

2.00076

OHCE' Si:P

g = 1.99890

as-grown + VUV20.4 GHz; 77K

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

Figure 2.21: Overall K-band (Pµ∼8 µW ; Bm∼0.6 G) ESR spectrum observedon the as-grown fumed silica sample after VUV irradiation. The OHCs clearlydominate the spectrum.

2.5.3 The OHC type centers

Observations

Figure 2.21 shows an overall K-band spectrum measured on the as-grownsample immediate upon VUV irradiation over an extended field range atliquid nitrogen temperature. It corresponds to the totality of observed ESRsignals. The most eye-catching part in this ESR spectrum, without doubtconcerns the left part of the spectrum, exhibiting broad resonance features.The position and shape of the broad ESR signal(s) points to OHCs [82].The ESR signal was investigated with great care in order to assess moredetailed information about the specific ESR features. The ESR resonanceswere monitored as a function of observational temperature, revealing somestriking variations. It became clear that the broad ESR line centered atgc∼2.0074 is the resultant of at least two resonant responses.

One of them, centered at gc=2.0074±0.0005, exhibits powder patternproperties, and could be quite well simulated (dashed curves in Fig. 2.22)in the temperature range ∼77-100 K consistently for K and Q-band obser-vations using a simple spin Hamiltonian of effective spin=1/2, with one setof g matrix principal values given as g1= 2.0020, g2=2.0078, and g3=2.0670.


as-grown + VUV

Si:P

POR

33.9 GHz; 100K

OHC

20.7 GHz; 77K

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

E'

20 G

20 G

2.0074

Figure 2.22: K-band (Pµ∼8 µW ; Bm∼0.6 G) and Q-band (Pµ∼0.17 mW ; Bm=1G) powder pattern ESR spectra of the as-grown sample after VUV irradiation.Extreme ESR measurement parameters are used to enhance the signals in the grange 2.02-2.0015 and to saturate the other signals (E′-like center, methyl radical)present. The broad resonances centered at g∼2.0074 are ascribed to OHC centers.The dashed curves represent spectra simulations; The principal g values used forboth spectra simulations are g1=2.0020, g2= 2.0078, and g3=2.0670.

The latter g3 is taken from Ref. [4], and in fact is kind of immaterial here.Interestingly, these ESR parameters are very similar to those pertaining tothe well-known POR (O3≡Si–O–O•) [4]. To get more conclusive evidenceseveral attempts were made to resolve the powder pattern peak correspond-ing with the g3-component. However, these attempts failed and the signalremained beyond ESR detection, likely due to the low signal intensity andline broadening. Inferred defect densities were up to (3.1±0.3)×1015 g−1 orabout 1×10−3 defects/nanoparticle. The presence of PORs in the nanopar-ticles, albeit in small densities, would indicate that fumed silica is somewhatoxygen rich, which would be consistent with the low density of E′ centers(O vacancies) encountered.

The other part of the broad resonance line could not be assigned conclu-sively. There is a possibility that this line also concerns some (other) type of


OHCs, possibly the NBOHC (O3≡Si–O•, characterized by [4] g1=1.9999,g2=2.0095, and g3=2.078). But this is more of a suggestion.

However, even in lack of full identification, we still can provide, as auseful number, an estimate of the upper limit of the density of all OHCspossibly present from the added intensities of all pertinent signals. Thedensity was determined for different VUV irradiation times in order toensure that most, if not all, centers present are indeed ESR activated (at-taining the density saturation level). This leads to a maximum number ofOHCs possibly present of (0.07±0.01) defects/particle, which means thatonly one in about every 14 particles may contain such a defect. Even if onetakes all ESR active defects observed in the as-grown sample immediateupon VUV activation [the OHC-type centers, the E′-type center and themethyl radicals (see later)] into account still no more than one in aboutevery 14 particles may house an ESR-active defect. The broad resonanceline which can possibly be assigned to the NBOHCs, is indeed by far themost intense of the observed resonances.

Another observation is that the broad resonance lines, and therefore theOHCs density, decreases drastically when subjecting samples to additionalheat treatment.

Discussion

These observations contrast with the conclusions arrived at by Altman etal., who studied the light absorption of silica nanoparticles [155]. Theirexperiments revealed a non-exponential behavior of light absorption in vis-ible light, which they attributed to the presence of a large concentrationof NBOHCs. They also suggested that the concentration of these defectsamounts to such levels that they may cause narrowing of the optical energygap through strong Coulomb disorder. The current observations, however,strongly disfavor this suggestion, at least in a statistically uniform picture,since for these defects to cause changes in the band gap there should be atleast more than one defect present in each nanoparticle. Of course, defectbunching cannot be excluded, i.e., a small minority of particles could con-tain several to many OHCs per particle, the remainder and major portionof the SiO2 particles then being OHC free. We would have two particlesubsystems in terms of OHC content. So, without contesting any of thevaluable sets of date presented in Refs. [163, 164, 155], the current resultswould suggest that the underlying physical basis for interpretation of someresults needs adaptation. Probably the explanation for the observed nar-rowing in optical band gap should rather be searched for in systematic


structural differences in the a-SiO2 network of the nanoparticles as com-pared to bulk SiO2.

Another point of discussion is Glinka et al.’s suggestion that theNBOHCsare formed by the process presented in Eq. 2.1 during heat treatment ofthe silica nanoparticles in air at Tan≥600 ◦C and that these defects domi-nate the PL spectrum when the sample is pretreated at Tan=900 ◦C [164].Yet, the broad ESR resonance line observed here, which can possibly be as-signed to some type of OHCs, on the contrary, decreases drastically uponadditional heat treatment. According to Eq. 2.1 not only the concentrationof NBOHCs but also the concentration of E′-type centers should increasesince they are formed in pairs. However, no such increase in E′ concen-tration could be observed either. These two observations make it quiteunlikely for the reaction described by Eq. 2.1 to play a dominant role. Infairness, it should be mentioned that the post-manufacture heat treatmentsapplied by Glinka et al. (2 h in air) differ from those applied in this research(∼1 h in vacuum), which, given the nano dimensions of the particles, maybe linked to the difference in result attained. This would require furtherinvestigation.

800 850 900 950 1000 1050 1100 1150 12000.0

0.5

1.0

1.5

2.0

2.5

upon VUV

[EX]

(1015

g-1)

Tan (°C)

Figure 2.23: The EX defect density upon VUV irradiation as a function of post-manufacture heat treatment in vacuum. The solid curve is merely meant to guidethe eye exposing the peaked behavior of the defect generation.


2.5.4 The EX center

With increasing Tan, another signal appears in increasing intensity. Theoriginating defect is conclusively identified as EX, a SiO2-specific cen-ter [68, 69]. In one view, it is pictured as a hole delocalized over 4 non-bridging oxygens grouped around a Si vacancy. The signal is observed fromTan≥960 ◦C on and is most intense [density up to (4.9±0.5) 1015 g−1] afterannealing at Tan∼1115 ◦C; the intensity generally drops upon VUV irra-diation (cf. Table 2.1). The EX density, measured after VUV irradiationas a function of Tan is shown in Fig. 2.23. A characteristic spectrum isshown in Fig. 2.24. The measured ESR parameters (g=2.0024, hf splittingA=16.4 G) are characteristic for the well known EX center. The EX centerhas been reported in various works as discussed in section 1.2.2.

2.5.5 The methyl radical

Results

A characteristic ESR spectrum (Fig. 2.25) consisting of 4 hf lines withintensity ratio 1:3:3:1 is additionally observed in the as-grown sample andin the samples annealed up to 1005 ◦C, measured immediate upon VUV

12120 12130 12140 12150 12160

Tan=1115 °C34.0 GHz; 100 K

16.4 G

2.0024 1.99891

Si:P

EX

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

Figure 2.24: Q-band ESR spectrum (Pµ∼0.3 µW ; Bm=1 G) of the EX centermeasured in the samples subjected to post-manufacture heating at Tan=1115 ◦C.


12025 12050 12075 12100 12125 12150

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

Tan = 850 °C33.9 GHz; 100 K

CH3

Si:P

1.99891

Figure 2.25: Q-band ESR spectrum (Pµ∼0.165 mW ; Bm=1 G) of the CH3 radicalobserved for the nanoparticles subjected to post-manufacture heating at Tan=850◦C. The dashed red curve represents the CH3 radicals spectrum simulation usingthe parameters g=2.0026 and A=23 G.

irradiation. This specific intensity ratio is readily explained as originatingfrom a spin system consisting of an unpaired electron (S=1/2) interactingwith three equivalent I=1/2 nuclei. Since the sample is of very high puritythe most likely candidates for the nuclei are protons. The measured ESRparameters (average g value gav=2.0026±0.0005; A=23±0.5 G) are indeedvery similar to those observed for the methyl radical [213]. The methylradical has previously been reported to be present in irradiated amorphousSiO2 [214], on silica gel surfaces [215], and in some irradiated high puritysynthetic silica [216]. It leaves no doubt that the observed ESR spectrumoriginates from the CH3 radicals.

The total intensity of the spectra was inferred by double numerical in-tegration of the hf component on the outer right side of the spectrum (thebest resolved hf component here, the others being perturbed by signal over-lap) and multiplying this intensity by 8. The density of the methyl radicalsis the highest, i.e., (0.047±0.005)×1015 g−1, for the sample annealed atTan=850 ◦C subjected to VUV. This sample was used for further analysis.The methyl radical density as a function of post-manufacture heating in


0 200 400 600 800 1000 1200

0.00

0.01

0.02

0.03

0.04

0.05

0.06

upon VUV[C

H3. ] (

1015

g-1)

Tan (°C)

Figure 2.26: The CH3 radical defect density upon VUV irradiation as a functionof post-manufacture heat treatment in vacuum.

vacuum is shown in Fig. 2.26.

Analysis

As the measurements were carried out carefully with reduced modulationamplitude to avoid overmodulation, it became clear that the line widths ofthe four components of the observed CH3 signals depend on the nuclear spinquantum number (MI). This observation can be interpreted as the resultof the tumbling of a spin system with anisotropies in the hf interaction andg matrix, which provides a relaxation mechanism dependent on MI [217].When ESR measurements where carried out at a temperature below 85K, the ESR lines broadened beyond ESR detection. This result suggeststhat the radicals ’freeze’ at low temperatures, substantiating that they areindeed tumbling when measured at a sample temperature of about 100 K.

The tumbling frequency can easily be calculated (see reference [215] andthe references therein). Following their analysis, the line width in terms ofinverse transverse relaxation time can be expressed as a sum of a secular,pseudosecular, and nonsecular part:

(T2)−1 = (T2)−1sec + (T2)−1

pseudosec + (T2)−1nonsec. (2.4)


-1.5 -1.0 -0.5 0.0 0.5 1.0 1.52.0

2.5

3.0

3.5

4.0

4.5

5.0

5.5

6.0

6.5

Line

wid

th (M

Hz)

MI

Figure 2.27: Line widths of the four hf components of the methyl radical spectrumas a function of corresponding nuclear magnetic quantum number. The dots arethe experimental points; the solid line represents the fitted curve based on Eq. 2.5(see text).

Since the nonsecular contribution to the line width is completely negligibleand the pseudo-secular contribution is negligible to a fair approximation(∼10%), the expression for the line width can be reduced to

(T2)−1 = a0 + a1MI + a2M2I . (2.5)

The constants can be evaluated as:

a0 =445

(∆γB)2τc +K, (2.6)

a1 =415b∆γBτc,

a2 =15b2τc,

where τ c is the correlation time, b the hf anisotropy, B the magnetic field,∆γ the g matrix anisotropy, and K contains other unspecified contributionsto the line width that are independent of MI .


The line widths were measured and least squares fitted by Eq. 2.5.It should be noted that the determination of the exact line widths wassomewhat hampered because of signal overlap. The measured line widthsand calculated fit are shown in Fig. 2.27. The inferred constants are:a0=3.656 MHz, a1=1.017 MHz and a2=0.259 MHz. If using the valueb=4.5 MHz, obtained by Heller [218] for the CH3 group in the radicalCH3C(COOH)2,as an estimate for the hf anisotropy of the methyl radi-cal, according to previous usage [216], the correlation time and thus thetumbling frequency can be calculated. In this way we find τ c=6.39×10−8

s, giving 1.56×107 s−1 for the tumbling frequency. The latter value is inbetween the value inferred for the methyl radical trapped in irradiated highpurity fused silica [216] (1.1×107 s−1) and the value obtained for the rad-ical adsorbed on the silica gel surface [215] (2×107 s−1). It seems that infumed silica the radical’s rotation is slightly more hindered compared tothe methyl radicals adsorbed on a silica gel surface, but less hindered bythe silica network as compared to the methyl radicals present in irradiatedfused silica.

It should be noted that in fumed silica the methyl radicals are possiblyless stable than the ones observed in irradiated fused silica. In the lattercase Friebele et al. still measured the same defect densities after keepingthe sample for two years in room ambient [216]. In fumed silica, however,the defect densities decreased beyond ESR detection already several daysafter irradiation.

The appearance of CH3 radicals is a general indicator of carbon and Hcontamination. This may not come as a surprise with respect to hydrogen,the abundance of which has been amply demonstrated previously by numer-ous techniques [160, 165]. However, although no carbon should be presentduring the fabrication of fumed silica, the nanoparticles are clearly C con-taminated since the methyl radicals are observed in the as-grown state.Apart from possible C impurification during manufacturing, another pos-sibility is C-contamination arising from surface adsorbance of C-containingspecies (e.g., CO, CO2) from room ambient. Given the particles large sur-face area aspect, the latter may appear most likely. For completeness, weadd that in the synthetic fused silica study using 60Co γ-irradiation [216],the CH3 radical formation is ascribed to radiodissociation of CH4 or COmolecules trapped in the silica network during silica formation.

2.6 Summary and conclusions 113

2.6 Summary and Conclusions

A systematic ESR study in conjunction with post-manufacture irradiationand/or annealing in vacuum in the range 850-1175 ◦C or in the presenceof SiO release in the range 1005-1205 ◦C of fumed silica nanoparticles,reveals for the first time the presence of ESR-active point defects. Theirgeneration and properties have been studied systematically by ESR as afunction of Tan and VUV treatment. Several point defects inherent to theSiO2 network are revealed, including E′γ , the O−2 ion, and the EX center.The methyl radical is also found to be present. Additionally, an as yetunidentified defect (LU2) is newly observed. Besides the identified ESRsignals, still others are observed who could not be spectroscopically iso-lated or identified. The presence of these defects and their ESR propertiescan provide important information on the structural and chemical nature ofthe fumed silica nanoparticles. In particular, using the intrinsic E′ defectsas local atomic probes, the atomic properties of the SiO2 network of thenanoparticles have been assessed. Some specific results include the follow-ing.

(1) The methyl radical is observed in the as-grown state and after post-manufacture baking in the range 850-1005 ◦C, immediately upon VUVirradiation. These •CH3 centers are a familiar indicator of trace carbonand hydrogen contamination. At T∼100 K, it has been inferred that theradicals tumble at a frequency of 1.56×107 s−1. Yet, they appear unstableas their ESR signal intensity decreases beyond ESR detection several daysafter the VUV activation.

(2) Evidence has been found for the presence of oxygen-associated holecenters (NBOHCs and PORs) in the as-grown samples after VUV irradi-ation. The dominant OHC type occurring appears to be the peroxy radical(O3≡Si–O–O•), which would indicate that fumed silica is oxygen rich.

(3) An as yet unidentified defect, which may be linked to specific networkaspects, is newly observed in the as-grown samples annealed at Tan=850-960 ◦C even without additional VUV treatment. The spectral propertieshave been studied in detail, aiming its identification. The ESR propertiesinclude: gc=2.00276±0.00005, closely axial symmetry, and g matrix prin-cipal values g‖=2.0041 and g⊥=2.0027. As detection of any associated hfstructure has so far failed, the atomic nature of the originating defect re-mains largely unknown. Yet, based on specific ESR properties the center


might possibly be ascribed to O2Si≡Si•, an E′-like center with chemicallymodified back bond arrangement, or to a carbon related defect center.

(4) The spectroscopic properties of the E′ signal have been studied withparticular interest since defect analysis can provide access to the nanopar-ticle structure on atomic level.

(4.1) Accordingly, changes in the zero crossing g value, line shape, andsignal intensity as a function of anneal temperature, observational temper-ature, and elapsed time period between VUV activation and ESR measure-ment as the sample was kept in room ambient were carefully monitored.

The combination of all the observations on the E′ signal hints at thepresence of two different systems of E′ centers with superposed ESR sig-nals. It is suggested that the core of the nanoparticles exhibits a morebulk a-SiO2 like structure while the surface region is of a different, morestrained nature. There appears a complex strain distribution. The E′ cen-ters present in the core regions are very similar to the common E′γ centerspresent in bulk a-SiO2. The E′ centers in the surface region, however, ex-hibit different properties. This can be seen as due to enhanced amorphousstate disorder and strain at the surface and in near surface layers. Anotherimportant factor is that the surface E′ system is possibly more apt to en-vironmental chemical and physical interaction (adsorption) impact. As thepost-manufacture temperature increases, sintering takes place decreasingthe specific surface area [153] and thereby possibly the number of surfaceE′ centers. The overall maximum E′ density is small compared to standardthermal a-SiO2 grown on Si, indicative of a high quality of the fumed silicaparticles in terms of occurring O vacancies: Only one E′ center is detectedin about every 1000 nanoparticles.

(4.2) In addition the influence of the presence of an (100)Si/SiO2 in-terface during post-manufacture heating at elevated temperatures on thefumed silica nanoparticles has been studied. Through the observation andanalysis of the primary 29Si hf splitting of the E′ centers located in thecore region of the nanoparticles atomic scale information has been derivedon the ring structure composition. The results demonstrate that the hfparameters are more sensitive to changes in the network structure than theg matrix.

The presumed interaction of volatile SiO at high temperatures in vacuumwith the SiO2(x) network leads to an increase in the density of E′γ centers,observed after VUV activation, which indicates that the fumed silica net-work is vulnerable for SiO attack. With SiO introduction Si enrichment


occurs leading to an increase in O vacancy (E′) density. This would bringthe fumed silica network closer to the compositional properties of bulk SiO2

with known superior quality. Yet, the SiO influence does not appear to beuniform throughout the nanoparticle, as the surface regions and the coreregion response differently to interaction with SiO molecules.

The detection of the primary 29 hf structure of the E′ centers locatedin the core region of the nanoparticles was enabled by virtue of the drasticincrease in E′ defect density resulting from the interaction of the fumedsilica network with volatile SiO at high temperatures. Consistent fittingof the primary 29Si E′ hf structure revealed a larger isotropic hf constantfor the fumed silica nanoparticles than for suprasil. As the hf parametersare linked with the O–Si–defect bond angle θ, this result indicates that θis larger as well for fumed silica than bulk SiO2. The latter result pointsto a more pyramidal defect structure for the E′ centers in the core regionof fumed silica particles. Moreover the observed increase in Aiso can belinked with a densification of the SiO2-matrix. According to previous in-terpretation, this densification of the core of the nanoparticles is possiblydue to a modification of the distribution of occurring ring structures in thefumed silica network suggesting the presence of more low-membered ringsof SiO4-tetrahedra.

Besides the primary 29Si E′ hf doublet a H-associated doublet assignedto O2≡Si•–H entities was additionally observed. Relatively to the ordinaryE′ signal, this doublet is found to be less prominent in the spectra of fumedsilica than suprasil indicating that the nanoparticles would contain less hy-drogen.

As a general conclusion, the obtained results illustrate that detailed ESRstudies of the specific ESR parameters such as the g and hf matrix of anembedded inherent point defect (such as E′) can provide valuable informa-tion on the specific network structure on atomic-scale –in the current case,of nanosized particles. As known, the fundamental strength derives fromthe local probing power of the unpaired electron’s wave function.

Chapter 3

P-associated defects in thehigh-κ HfO2 and ZrO2insulators revealed by ESR

3.1 Introduction

In this chapter as well as in the next, we will step into the field of high-κresearch. As aforementioned (section 1.3), excessive leakage currents en-countered in metal-oxide-semiconductor field-effect transistors (MOSFETs)as the conventional a-SiO2 gate dielectric is scaled down to the 1-nm rangewill require the replacement of the gate insulator by one of higher κ. In thischapter we will focus on ZrO2 and HfO2, two prominent high-κ oxides. Thelatter is currently considered as one of the most promising candidate high-κ dielectrics because of its favorable properties including thermal stability,suitable barrier heights for electrons and holes, and advantageous electri-cal properties, in part due to the existence of an ultrathin SiO2 interlayerbetween the high-κ film and Si [20].

Nevertheless, to serve as a viable candidate the alternate dielectric muststill meet various other stringent requirements, some of them not yet fullyexplored. One of these is that the high-κ oxide should display superb re-sistance to dopant penetration during necessary dopant activation anneals,as such penetration would detrimentally affect device performance. Thus,to be successful the alternate dielectric should be an excellent diffusionbarrier, and consequently, the behavior of a dopant such as phosphorus isof key interest. A few works have reported on the diffusion of dopants inhigh-κ dielectrics [219, 220, 221]. Alarmingly, with respect to HfO2-based

117

118 Chapter 3: P-associated defects in HfO2 and ZrO2

dielectrics, the results indicated an enhanced penetration of P through HfO2

or Hf-silicates compared with standard SiO2 films.

3.2 Phosphorus diffusion

Defect sites resulting from P atoms incorporated in the gate oxide may actas charge traps degrading device performance. Consequently, the presenceof P atoms in the network of the gate dielectric is something one should tryto avoid. Most unfortunately, this appears quite impossible. For instance,during device processing the MOSFET’s source and drain are implantedwith a high density of P atoms (∼1020 at. cm−2). It is not unlikely thatduring implantation part of the P atoms are scattered into the sides of theneighboring gate oxide. Moreover, the thermal steps necessary for dopantactivation in device manufacturing may result in the diffusion of P atomsfrom the P-doped source and drain or, when no metal gate is used, fromthe P-doped poly-crystalline Si (poly-Si) gate into the high-κ dielectric.

Consequently, the problem of P penetration has been studied by severalresearch groups [219, 220, 221]. Quevedo-Lopez et al. studied penetra-tion of P through HfSixOy and HfSixOyNz films in stacks consisting of aP-implanted (5×1015 at.cm−2) poly-Si film deposited by CVD on a CVDHfSixOy (5 nm) or HfSixOyNz film on (100)Si [219, 220]. From the sec-ondary ion mass spectroscopy (SIMS) P depth profiles in the Si substrate

Figure 3.1: SIMS P depth profile in the Si substrate after 1000 ◦C annealing for60 s and chemical etching of the P-doped poly-Si/HfSixOy (HfSixOyNz)/Si stack.The dotted curve indicates the profile for the P-implanted as-grown stack. Thefigures is taken from Ref. [219].

3.2 P diffusion 119

after rapid thermal annealing in the range 900-1050 ◦C, such as the oneshown in Fig. 3.1 for Tan∼1000 ◦C, the authors concluded that P pene-trates through the HfSixOy layer into the Si substrate for Tan≥1000 ◦C.The P penetration through the HfSixOy films upon annealing at these tem-peratures was shown to be substantially higher than for HfSixOyNz films,where it was near the detection limit. The P diffusivity in the HfSixOy

films is found to be at least one order of magnitude higher than those inSiO2 and SiOxNy.

Based upon the film microstructure, studied by HRTEM, the authorspropose that this enhanced P diffusion in HfSixOy is related to grain bound-ary formation resulting from HfSixOy film crystallization. Fig. 3.2 (a) showstheir HRTEM result for a P-doped poly-Si/HfSixOy/Si stack subjected to a60 s rapid thermal anneal at 1050 ◦C, clearly displaying the crystallizationof the Hf silicate film after annealing. The reduction of dopant penetrationthrough the introduction of N into the HfSixOy films is linked with thesuppression of crystallization observed in the HfSixOyNz films [see Fig. 3.2(b)]. This is attributed to both the lower Hf content in these films and theN incorporation.

Suzuki et al. studied the P diffusion in thick (110 nm) P-implanted(5×1013 at. cm−2) HfO2 films deposited by ALCVD on Si, upon rapidthermal annealing in N2 [221]. From the SIMS profiles of the P concentra-tion in HfO2, such as shown in Fig. 3.3 (a) for Tan∼1100 ◦C, the authorsextracted the diffusion coefficient (Fig. 3.3 (b)). They concluded that P

(a) (b)

Figure 3.2: HRTEM results for 60 s at 1050 ◦C annealed stacks of (a) P-dopedpoly-Si/HfSixOy/Si (the red ellipse circles the crystalline region) and (b) poly-Si/HfSixOyNz/Si (no detectable crystalline region is observed). The figures aretaken from Ref. [220].


(a) (b)

Figure 3.3: (a) SIMS profiles of the P concentration in as-grown P-implantedALCVD HfO2 on Si and after rapid thermal annealing at 1100 ◦C. (b) Diffusioncoefficients of As and P in HfO2, and B in HfO2, Si, and SiO2. The figures aretaken from Ref. [221].

diffusion through HfO2 is significant –two orders of magnitude higher thanthat of B in SiO2.

Thus rather than solely focussing on finding ways to avoid P incorpora-tion into the gate dielectric, it might be of vital importance to additionallyfocus on the assessment of the electrical activity of the defect sites resultingfrom this P penetration into the high-κ oxide. To reach this goal, atomicscale information will be required of how dopants are incorporated into thehigh-κ network. For this purpose ESR has been demonstrated to be thepreferred technique as evidenced by the reliable identification of numerous(impurity related) point defects in various dielectrics, such as SiO2. Onepertinent class here concerns dopant-associated centers [4]. Indeed, ESRstudies of P-associated defects in SiO2-based glasses resulted in the obser-vation and identification of several defect centers. This may be evidentfrom the overview given in section 1.2.3 of the different P-associated de-fect centers so far observed in SiO2-based glasses [61, 92, 13, 87, 88] andP-doped c-SiO2 [85, 86].

The present chapter deals with the ESR observation of P-related pointdefects in two different oxides eminent in the current high-κ dielectric re-search, i.e., ZrO2 and HfO2. In each case, a similar type of defect is ob-served, the results indicating that the P atoms can be substitutionallyincorporated in both the ZrO2 and HfO2 network resulting in ESR activedefect centers exhibiting similar ESR parameters. In both oxides a 31P hf


doublet of large splitting is observed, assigned to a P2-type defect. Thebasic parameters of the defect are assessed to retrieve the defect’s atomicnature which might prove essential in understanding their possible chargetrapping behavior, and hence, detrimental effect on device performance.


3.3.1 Samples

A first set of samples consisted of 99.96% pure ZrO2 powder of particles size<44 µm and a density of 5.89 gcm−3 obtained from Alfa Aesar. As mainmetallic impurities the powder contains <50 ppm Hf and <25 ppm Al, Fe,P, Si, and Ti. The particles are crystalline and in the monoclinic phase.After initial observation of P-associated centers by ESR a second set ofsamples was obtained, invoking another prominent high-κ oxide, throughdepositing 100-nm thick amorphous HfO2 on p-type (100)Si substrates,using ALCVD at 300 ◦C from HfCl4 and HfO2 precursors. Inherent to thisfabrication method is that the HfO2 films contain 2-3% H. To boost ESRdetectivity the samples were implanted by P ions to densities of ∼ 1015

cm−2, with the implantation energy adjusted to attain a mid-film meanprojected range.


Initial conventional cw absorption derivative Q-band (∼34 GHz) ESR mea-surements were performed at 100 K on the ZrO2 powder. The modulationamplitude (Bm) of the applied magnetic field and the incident microwavepower (Pµ) were restricted to levels not causing (noticeable) signal dis-tortion. A co-mounted Si:P marker sample [g(100 K)=1.99891] or a Li:Fmarker sample (g=2.00229) was used for g factor and (spin) density cal-ibration. The latter was attained through orthodox double numerical in-tegration of the recorded first-derivative absorption spectra of the defectand the marker. However, to circumvent saturation, in this way increas-ing sensitivity, the rest of the ESR measurements were performed in thesecond harmonic mode [14, 15] in X (∼9.2 GHz), K (∼20.4 GHz), andQ-band in the temperature range 4.2-300 K using relatively high modula-tion amplitudes and microwave powers. As outlined elsewhere, it appearsthat when appropriate conditions are met for a certain defect center, thesecond harmonic phase-quadrature (out-of-phase) spectra are identical tothe absorption spectra of that defect [13, 14, 15].


Separate sets of the HfO2 samples were subjected to post-implantationannealing in N2 for ∼30 min at desired temperatures (Tan) in the range 300-900 ◦C. After initial ESR tests, to maximally reveal all defects, the sampleswere subjected to suitable photon irradiation. All samples were subjectedto unbiased VUV irradiation (10 eV photons; flux∼ 1015 cm−2s−1) ob-tained from a Kr-resonant discharge lamp. The nm-thin HfO2 films wereonly irradiated for a relatively short period (20 min), while the µm-sizedZrO2 particles needed to be irradiated for days due to the limited pene-tration depth of the VUV photons and the shadowing effect with powderparticles. To overcome the latter the powder was regularly stirred duringirradiation. A separate set of ZrO2 samples was subjected to UV irra-diation (∼2-7 eV , obtained from a Xe-lamp) for several hours. DuringUV irradiation the powder was contained in a continuously rotating quartzholder put in front of the lamp. Possibly, the treatment may additionallyunveil strained or weak bonding (bond rupture) and activate diamagneticprecursor sites [112].

3.4 Results

Curve (a) in Fig. 3.4 shows a typical high-power second harmonic mode [14,15] Q-band (∼34 GHz) spectrum observed in 99.96% pure ZrO2 powder ofparticle size <44 µm subjected to UV irradiation. Various signals appearbut here we will only focus on the pair of resonance features (doublet) la-beled P2 in Fig. 3.4. From careful analysis of the saturation behavior of thedifferent resonances, it was established that the resonance lines observed inthe center region of the spectra have an origin different from the resonancelines labeled P2. The doublet was observed in the as-received particles butbecame much more prominent upon VUV or UV irradiation. The doubletcould be detected in both the high-power second harmonic mode [Fig. 3.4

curve (a)] and the conventional low-power derivative-absorption mode [222][Fig. 3.4 curve (c)]. After taking the first derivative of the second harmonicmode spectrum [Fig. 3.4 curve (b)], identical spectra are obtained in bothmodes, attesting that both modes can reliably be used to obtain correctESR parameters (such as, g matrix, hf tensor, and line width) of the ob-served defects. Additionally, the second harmonic mode measurements ofthe UV irradiated ZrO2 powder were performed at various (measurement)temperatures T in the range 25-297 K. The observed doublet spectrumdid not change indicating that the ESR parameters are not influenced bythe observation temperature.

3.4 Results 123

11250 11500 11750 12000 12250 12500 12750

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

(c)

(b)

ZrO2 + UV~33.9 GHz; 100 K

P2

(a)

Figure 3.4: Q-band ESR spectra of ZrO2 powder subjected to UV irradiation mea-sured at 100 K: (a) the high-power (saturation) second harmonic mode spectrum,(b) the first derivative of the high-power (saturation) second harmonic mode spec-trum, and (c) the conventional low-power derivative-absorption mode spectrum.

A similar doublet spectrum was observed in the P-implanted HfO2 filmson Si but only after subjecting the sample to post-implantation annealingin the range 500-900 ◦C followed by VUV irradiation. This is illustratedin Fig. 3.5 [curve (b)] showing a representative Q-band spectrum for theHfO2 sample subjected to post-implantation annealing at 900 ◦C, the tem-perature for which the doublet became the most prominent, and additionalVUV irradiation. Moreover, Fig. 3.5 illustrates that the resonance linesobserved in the ZrO2 powder [spectrum (a)] and in the HfO2 films arequite similar and, even though the ESR parameters slightly differ, seem tooriginate from similar defect centers.

Based on the general knowledge of powder pattern ESR line shapes, thekind of symmetry relationship between the two resonance features in one


11250 11500 11750 12000 12250 12500 12750 13000

magnetic field (G)

(b)dPµ/d

B (a

rb. u

nits

)

Q-band

P2

(a)

Figure 3.5: First derivative of the high-power (saturation) second harmonic modeQ-band spectra (a) measured at 33.91 GHz at 100 K in ZrO2 powder subjectedto UV irradiation and (b) measured at 34.03 GHz at 4.2 K in P-implanted HfO2

films on Si subjected to a 900 ◦C post-implantation anneal in N2 and subsequentVUV irradiation. The dashed red curves represent the simulations obtained usingthe principal g matrix values (a) g1=2.0011, g2=2.0007, and g3=2.0007 and (b)g1=1.9965, g2=1.9975, and g3=1.9975, the principal hf tensor values (a) A1=1370G, A2=1162 G, and A3=1157 G and (b) A1=1425 G, A2=1245 G, and A3=1160G, a Gaussian line shape, and peak-to-peak line widths (a) ∆B1

pp=8 G, ∆B2pp=17

G, ∆B3pp=17 G and (b) ∆B1

pp=24 G, ∆B2pp=18 G, ∆B3

pp=24 G.

3.4 Results 125

11250 11500 11750 12000 12250 12500 12750 13000

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

(a)

6500 6750 7000 7250 7500 7750 8000 8250

(b)

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

2200 2400 2600 2800 3000 3200 3400 3600 3800 4000 4200

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

(c)

Figure 3.6: The high-power (saturation) second harmonic mode Q, K, and X-band spectra observed in P-implanted HfO2 films on Si subjected to a 900 ◦C post-implantation anneal in N2 and subsequent VUV irradiation. The dashed red curvesrepresent the simulations obtained using, for all three spectra, the principal g ma-trix values g1=1.9965±0.0004, g2=1.9975±0.0004, and g3=1.9975±0.0004 andthe principal hf tensor values A1=1425±10 G, A2=1245±10 G, and A3=1160±10G.


spectrum would leave little doubt about their correlated nature. Yet, tofirmly establish whether the two resonance lines do form a natural doublet,complementary X (∼9.2 GHz) and K-band (∼20.4 GHz) ESR experimentswere carried out, as illustrated by the spectra shown in Fig. 3.6 and Fig. 3.7for the case of (100)Si/HfO2 and ZrO2, respectively. It appears that thedoublet field splitting slightly increases (∼40G) from Q to X-band measure-ments, but is in first order independent of the observational frequency. Thecentral g value (gc) of the doublet, however, shifts significantly downward(∆gc∼0.064) from X to Q-band (cf. Fig. 3.7). This shift in gc is found

~ 9.2 GHz

~ 20.4 GHz

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

300 G

ZrO2 + UV

~ 33.9 GHz

P2

g~2.005

Figure 3.7: First derivative of the high-power (saturation) second harmonic modeQ, K, and X-band spectra observed in ZrO2 powder subjected to UV irradiation.The dashed red curves represent the simulations obtained using the principal gmatrix values g1=2.0011±0.0004, g2=2.0007±0.0004, and g3=2.0007±0.0004 andthe principal hf tensor values A1=1370±10 G, A2=1162±10 G, and A3=1157±10G. The spectra are aligned at the resonance field of a main central signal atg∼2.005 revealing the significant shift of the doublet as a function of observationalfrequency.

3.4 Results 127

to be consistent with the Breit-Rabi shift [18] expected for a hf doubletresulting from a S=1/2, I=1/2 system exhibiting such large hf splitting(∼1200 G), as illustrated in Fig. 3.8 for the doublet observed in ZrO2:The experimentally obtained gc values (filled symbols) for three differentmicrowave frequencies match within experimental error the variation in gcvalues calculated using exact matrix diagonalization for a S=1/2 and I=1/2spin system with A=1160 G and g=2.0007. Furthermore, counter to thecommon spectral appearance of S>1/2 centers, no ∆MJ=2 transitions athalf magnetic field strength of the ∆MJ=1 transition could be observed.

10 15 20 25 30 352.00

2.01

2.02

2.03

2.04

2.05

2.06

2.07

2.08

g c

Microwave frequency (GHz)

Figure 3.8: Zero-crossing g values (gc) of the measured P2 signal in ZrO2 pow-der subjected to UV irradiation for three different microwave frequencies. Filledsymbols: experimentally obtained gc. Open symbols: calculated gc obtained usingexact matrix diagonalization for a S=1/2 and I=1/2 spin system with A=1160 Gand g=2.0007. The solid line is merely meant to guide the eye showing the upwardshift of the gc with decreasing microwave frequency.


3.5 Analysis

Altogether these observations leave little doubt that the two signals concerna hf doublet resulting from interaction of an unpaired electron of spin S=1/2with a closely 100% naturally abundant nucleus of spin I=1/2, rather thanstemming from two dissimilar paramagnetic species. In looking for a can-didate I=1/2 nucleus of 100% natural abundance which could evoke suchlarge hf splitting (∼1200 G), i.e., large nuclear moment, a quick perusalof the table of isotopes would leave only two possibilities, i.e., 31P or 1H.Taking into account the presence of P impurities in non-negligible amounts(<25 ppm) in the ZrO2 powder and the P-implantation ([P]∼1015 cm−2)signal boost in the HfO2 samples, 31P follows as the most likely candidatenucleus involved.

3.5.1 Spectra simulations

The observed doublet spectra, exhibiting typical powder pattern proper-ties, could be consistently described (cf. Fig. 3.5, dashed curves) by oneeffective spin S=1/2 and 100% naturally abundant I=1/2 center, accordingto the simplified spin Hamiltonian comprised of the Zeeman and hf terms,respectively

HS = µB ~B · g · ~S + ~S · A · ~I. (3.1)

For each of the dielectrics separately, the fitting, using a code based onexact matrix diagonalization incorporating the Breit-Rabi formula, couldbe consistently performed with one set (within experimental error) of g andA principal values for the independent X, K, and Q-band spectra, provid-ing high confidence in the inferred data. The fitting result is illustrated inFigs. 3.6 and 3.7 (dashed curves) for the (100)Si/HfO2 and ZrO2 case, re-spectively. The inferred effective S=1/2 Hamiltonian parameters are listedin Table 3.1. The indicated experimental accuracies on the inferred g and Avalues are dominated by the accuracy attained on the magnetic field sweepsince for the wide scans (>1000 G) the magnetic field sweep is not perfectlylinear. Measurements at three different microwave frequencies and thus inthree different experimental setups, however, resulted not only in more reli-able spectra simulations but also in a reduction of the derived experimentalerror on the ESR parameters.

From the inferred parameters listed in Table 3.1, it may be noticedthat the hf doublets observed in ZrO2 and HfO2 are quite similar, stronglysuggesting that both may originate from nominally the same defect centersimilarly embedded in a yet different oxide network. Quite naturally, as

3.5 Analysis 129

well expected for non-identical environment, there will emerge some (small)differences in the obtained ESR parameters as can be seen from Fig. 3.5and Table 3.1. Compared to the HfO2 case, the defect center observed inthe ZrO2 powder exhibits ESR parameters closer to axial symmetry, the hfsplitting is smaller, the principal g values are closer to the free electron gvalue, and the line widths are smaller.

3.5.2 Structural model

Having established pertinent ESR parameters and accepting the P-relatednature of the observed defect centers the next step in analysis would at-tempt to assess the structural model. In that search, let us be guided bya comparison of our results with ESR data of P-associated defects previ-ously detected in other oxides, in particular those in doped silica glasses[4, 61, 92, 13, 87, 88]. Characteristic for the defects observed in the currentstudy is the very large hf splitting, a property only observed for two typesof P-associated defects in SiO2 based glasses, i.e., P1 and P2. Comparingour g and hf data with those from literature (cf. Table 1.3 and 3.1) thecurrently observed defects appear to be most similar to the P2 defect (ex-hibiting the largest hf splitting) observed in, e.g., P2O5-SiO2 glass [61] andP-implanted α-quartz [85, 86], assigned to a P substituting a Si atom. So,mainly based on a comparison of the Hamiltonian parameters, we assignthe currently observed defect centers in HfO2 and ZrO2 to P2-type defects.

In SiO2 the P2 center has been ascribed to a P substituting a Si atom:A phosphorus atom of formal oxidation step +5 has taken the place ofa silicon atom of oxidation state +4, hereby introducing a ”precursor”defect that provides a coulombic trapping potential for an electron. Thismodel was confirmed by an ESR study of single-crystal P-doped α-quartz.[85, 86] Based on the observed symmetry properties of the defect it wassuggested that the P2 defect consists of a P atom back bonded to 4 O atoms,with the P nucleus considerably displaced. As mentioned, this model ofsubstitutional P in α-quartz has recently been confirmed by theory [89, 90].

3.5.3 LCAO analysis

Proceeding with the analysis of the observed P2-type defects, it might beinteresting to determine the s (α2) and p character (β2) of the orbitalcomposition of the unpaired electron and its localization (η2) on the centralP atom [4]. To ease comparison we will do this in accordance with previousstudies of P2 in SiO2, using the LCAO approximation elaborated on in


Table

3.1:C

omparison

ofthe

ESR

parameters

ofthe

P2

defectem

beddedin

different

oxides.E

stimated

accuracieson

inferredg

andA

valuesin

thisw

ork([a],

[b])are±

0.0004and±

10G

,respectively.

g1

g2

g3

A1

A2

A3

α2

β2

η2

(G)

(G)

(G)

c-HfO

2[a]

1.99651.9975

1.99751425

12451160

0.330.67

0.83c-Z

rO2

[b]2.0011

2.00072.0007

13701162

11570.33

0.670.79

c-SiO2

[c]2.0012

2.00321.9991

12291087

10750.39

0.610.61

a-SiO2

[d]2.001

2.0012.001

13001150

11500.40

0.600.63

[a]P

2in

P-im

plantedH

fO2

subjected

topost-im

plantationannealing

inthe

range500-900

◦Cand

VU

V(10

eV)

irradiation.[b]

P2

inZ

rO2

powder

with

<25

ppm

Psub

jectedto

VU

V(10

eV)

orU

Virradiation

(seetext).

[c]P

(I)variant

ofP

2inα

-quartzsub

jectedto

x-rayirradiation

(seeR

ef.[85,

86]).[d]

P2

inP

2 O5 -SiO

2glass

subjected

toγ-ray

orx-ray

irradiation(see

Refs.

[4,61]).

3.5 Analysis 131

section 1.1.1. The results of such a LCAO analysis are also compared inTable 3.1 together with data on the P2 defect in silica based glasses andquartz compiled from literature. In each case the atomic coupling constantscalculated for the neutral 31P atom by Morton and Preston [8] were usedas those are currently regarded as the most accurate. It should be noted,however, that in previous works different values for As and Ap were usedresulting in different inferred values for α2 and β2 than the ones reportedin Table 3.1.

From the comparison, it appears that the main difference concerns thelocalization: the unpaired electron of P2 in HfO2 or ZrO2 is stronger (∼31%)localized on the central P atom than in SiO2. The unpaired electron or-bital composition, however, is also different: the proportion of the p to the scharacter has increased (β

2

α2 ∼2) for P2 in the high-κ oxides compared to P2

in SiO2 (β2

α2 ∼1.5). Some structural variations are of course expected whensubstitutionally embedding P in different matrices (environments), reflect-ing in the unpaired electron sp hybrid which will result in modification ofg and A values.

3.5.4 Defect densities

Next, we may wonder about the fraction of P impurity atoms that ulti-mately end up as an ESR-active P2 defect. For the P2 defects in ZrO2 anaccurate defect density could be obtained as the centers could be detectedusing conventional first-derivative ESR, giving, after prolonged UV irradi-ation a density of ∼1×1015 g−1, which means that at least 1% of the Pimpurities (<25 ppm, as specified) results in an ESR active point defect.

The experimental situation is less favorable for the (100)Si/HfO2 case asdue to excessive saturation in combination with the limited amount of sam-ple vailable, the P2 signal could not be correctly measured (unacceptablesignal distortion present or signal not observable at all) using conventionalESR. Quantitative determination of defect densities from second harmonicsaturation spectra, however, is not straightforward [13] and requires ex-treme care. The amplitude of the second harmonic mode spectra dependsin a complex way on the relaxation times of the defect centers and onvarious experimental parameters such as the modulation amplitude, themodulation frequency, and the microwave power [13].

Only via an indirect calibration procedure linking the correct first-deriva-tive low-power ESR spectra of P2 in ZrO2 with the high-power second har-monic saturation spectra, an estimate could be made of the P2 density inthe HfO2 samples along the following procedure: For both the ZrO2 and


the HfO2 samples a second harmonic Q-band spectrum was recorded of thedoublet using the same experimental parameters. Each second harmonicmeasurement was immediately followed by a conventional first-derivativemeasurement of the co-mounted marker again under the same experimen-tal conditions for both samples. This way the quotients can be determinedof the intensity of the doublet, obtained through single numerical integra-tion of the second harmonic spectrum, to the intensity of the marker, ob-tained through double numerical integration of the first-derivative markerspectrum. Assuming that the relaxation times for P2-type centers in bothhigh-κ dielectrics are comparable, the ratio of the quotients will be equal tothe ratio of the total number of P2-type defects in the two materials. Know-ing the P2 density in the ZrO2 sample, this resulted in a total of ∼4×1013

and ∼9×1012 P2 defects in the ZrO2 sample subjected to UV irradiationand in the HfO2 sample subjected to post-implantation annealing at 900◦C and VUV irradiation, respectively. It appears that the total number ofP2 defects actually in the cavity when measuring the HfO2 sample is morethan 4 times lower than when measuring the ZrO2 sample, accounting forthe lower signal to noise ratio of the HfO2 spectra (cf. Figs. 3.5, 3.6,and 3.7).

In this way, for the HfO2 sample subjected to a post-implantation annealat 900 ◦C and VUV irradiation an estimated defect density of ∼5×1012

cm−2 was obtained, which would indicate that about ∼0.5% of the im-planted P atoms results in a P2 defect –a fraction quite comparable withthe ZrO2 case and P2 in atomic pressure CVD deposited BPSG films on Sisubjected to VUV irradiation, electron, and hole injection [13]. It shouldbe noted, however, that the inferred defect densities probably concern justlower limits as the VUV or UV irradiation applied here might not havebeen completely exhaustive in activating precursor sites into the ESR ac-tive state.

3.6 Discussion

In the HfO2 samples the P2 center is only observed after post-implantationannealing at Tan≥500 ◦C and additional VUV irradiation. The tempera-ture of the necessary ”turn on” post-implantation anneal step is close to theknown onset temperature for crystallization of the HfO2 film into a mixtureof tetragonal and monoclinic phases, with, notably, the monoclinic phasestrongly dominating [223, 224]. Hence, we suggest that the crystallizationis accompanied by part of the implanted P atoms then taking the required

3.6 Discussion 133

c

ba

a=b≠ca^b=120°

a

bc

a=b≠c

a

bc

a ≠ b≠ca^c=99.18°

Si

O

Zr/Hf

O

O

Zr/Hf

Figure 3.9: Illustration of the unit cell of the lattice structure of tetragonal HfO2

or ZrO2.

c

ba

a=b≠ca^b=120°

a

bc

a=b≠c

a

bc

a ≠ b≠ca^c=99.18°

Si

O

Zr/Hf

O

O

Zr/Hf

Figure 3.10: Illustration of the unit cell of the lattice structure of monoclinicHfO2 or ZrO2.


substitutional position in a particular crystalline phase. Information re-garding the latter may be inferred from the fact that the P2 center was alsofound to become more prominent upon annealing at higher temperatures.Indeed, it has been shown that the volume fraction of the tetragonal phasein the HfO2 film decreases with increasing temperature [223], the film al-most completely converting into the monoclinic phase upon annealing atTan∼900 ◦C. Apparently the P2 defect preferentially appears in the mon-oclinic phase. Figure 3.9 and 3.10 illustrate the unit cell for the tetragonaland monoclinic phase of HfO2 or ZrO2, respectively. For the ZrO2 powderno post-implantation anneal step was necessary to observe P2. This resultis still consistent within the above picture as the as-received ZrO2 particlesare in the monoclinic crystalline phase, where the P impurities were in-corporated during high-temperature particle formation, during which (partof) the P atoms could take a Zr substitutional position.

We thus have reached the result that the P2 defects observed in bothhigh-κ oxides studied are very similar and both pertain to the monoclinicoxide phase. Looking deeper, these findings seem well founded as the struc-tural parameters for m-HfO2 and m-ZrO2 are almost identical (cf. Ta-ble 3.2) [225]. Quite naturally, the small differences in the hf parameters ofP2 in both oxides just reflect the small differences in the structural param-eters. Comparing, however, our results with those for P2 in c-SiO2 [85, 86],the hf parameters were found to deviate more substantially. This can beseen as originating from the different network structure of the α-quartz,

c

ba

a=b≠ca^b=120°

a

bc

a=b≠c

a

bc

a ≠ b≠ca^c=99.18°

Si

O

Zr/Hf

O

O

Zr/Hf

Figure 3.11: Illustration of the unit cell of the lattice structure of trigonal α-quartz.

3.6 Discussion 135

Table 3.2: Comparison of experimental structural parameters for the monoclinicphase of chemically prepared HfO2 and ZrO2 powder, and for the trigonal phaseof SiO2 (e.g., α-quartz). Lattice parameters a, b, and c are in A.

a b c monoclinic angle βm-HfO2 [a] 5.1156 5.1722 5.2948 99◦11’m-ZrO2 [a] 5.1454 5.2075 5.3107 99◦14’t-SiO2 [b] 4.92 4.92 5.41 /

[a] Values taken from Ref. [225]. The accuracy on a, b, and c is ±0.0005. Theaccuracy on β is ±0◦05’.[b] Values taken from Ref. [226]. The accuracy on a, b, and c is ±0.01.

i.e., trigonal phase as illustrated in Fig. 3.11 [226]. A steric model for theadduced P2-type defect in HfO2 is depicted in Fig. 3.12. Thus, the P2-type defect may seem ubiquitous to all binary oxide glasses, where the finedetails of difference in structure may be addressed by theory.

After post-implantation heat treatment at Tan∼500 ◦C the P2 centersin the HfO2 samples were still found to reside in an ESR-inactive state(or at least, too low an ESR active fraction). Irradiation by VUV wasrequired to observe the hf doublet. In the ZrO2 powder this activation stepwas not necessary but additional VUV or UV irradiation did drasticallyincrease the defect density. A possible reason for the photonic impact isthat (part of) the P-defects are ESR inactive due to the presence of a chargecompensator [85, 86]. The VUV or UV irradiation would then cause chargetransfer resulting in the proper ESR active state. However, no evidence wasfound for the presence of such a charge compensator. A second possibility isthat (part of) the centers are left ESR inactive due to binding with H. TheVUV or UV photons would then photodissociate H from the passivateddefects [112]. The latter scenario is quite likely, especially for the HfO2

samples as these films contain 2-3% H [227].Having provided evidence that for both high-κ oxides a sizeable fraction

(∼0.5-1%) of the P atoms results in a P2 defect, we should draw atten-tion to their potential operation as charge traps, which would make themdetrimental for device performance [220, 221, 13]. There seems little doubtabout this: In SiO2 it has been concluded theoretically as well as experi-mentally that the P2 defects act as a hole trap [87, 93]. Whether the P2

136 Chapter 3: P-associated defects in HfO2 and ZrO2 P2 in HfO2

Hf/Zr

O P + e-

Figure 3.12: Proposed conceptual model and the formation of the ESR-active P2

defect in ZrO2 and HfO2, formally adopting 2-fold coordinated O atoms.

defects observed in the current high-κ oxides will also act as hole traps de-pends on the position of the energy level of the defects in the band gap ofthe oxides. For this theory could provide valuable information to eventuallyestablish the charge trapping behavior of the P2 defects embedded in HfO2

and ZrO2 [228]. The matter becomes even more crucial as recent studiesindicate enhanced P diffusion through HfO2 compared to SiO2 [220, 221].

3.7 Conclusions

In summary, we have reported on the observation by ESR of phosphorusrelated point defects in two prominent high-κ oxides, i.e., ZrO2 and HfO2.These donor related defects exhibit similar ESR parameters and are bothassigned to a P2-type center, largely based on comparison with the wellknown P-associated defects in silica. Thus the defect is ascribed to a P4+

atom substituting for Zr or Hf in the monoclinic phase of the ZrO2 andHfO2 matrix, respectively, where based on elemental LCAO analysis, theunpaired electron is found to be strongly localized on the P atom. A sizeablefraction of the incorporated P impurities was found to result in ESR activeP2-type defects. As these may act as detrimental charge traps in the oxide,the results of recent studies pointing out enhanced P diffusion throughHfO2 compared to SiO2, may urge one to establish the latter into moredetail to evaluate the potential effect on MOS-based device operation. This

3.7 Conclusions 137

will require correlative studies combining ESR with electrical and opticalmethods, backed up by theoretical insight.

In search of the replacement of the standard SiO2 gate insulator in MOS-FETS by a high-κ dielectric such as HfO2, one major obstacle encountered isthe generally enhanced trap density in the metal oxide based layer. Worseeven, (enhanced) migration of dopant impurities resulting in additionaltraps during necessary dopant-activation thermal steps may add one moreelement to this concern. The current work represents an initial step inatomically identifying potential charge traps in favored high-κ oxides fordevice application.

Chapter 4

Paramagnetic point defectsin (100)Si/LaAlO3structures: nature andstability of the interface

4.1 Introduction

In the previous chapter we focussed on those high-κ oxides that will replacethe conventional SiO2 gate oxide in order to overcome present day and nearfuture scaling issues. However, due to the inevitable presence of an SiO2-type interlayer as discussed in section 1.3.1, the implementation of HfO2

or nitrided hafnium silicates will only provide a temporary solution. Thepresence of such a ’low-κ’ interfacial SiOx layer in Si/high-κ structures, eventhough possibly providing the scope of realizing standard Si/SiO2 interfacequality, increases the net EOT. If the interlayer is pure SiO2 its physicalthickness (dSiO2) is added to the net EOT:

EOTnet = EOThigh−κ + dSiO2 , (4.1)

putting severe restrictions on the physical thickness of the high-κ dielec-tric. For example, to obtain a net EOT of 1 nm, the presence of an SiO2

interlayer of 0.5 nm mandates a 50% decrease of the high-κ oxide physicalthickness.

Optimization of the EOT would thus instruct to look for a Si/high-κ insulator system without such an SiOx interlayer. And of course, it is

139

140 Chapter 4: nature and stability of the (100)Si/LaAlO3 interface

hoped that this can be achieved hand in hand with simultaneously realizingan Si/high-κ layer interface of superb electrical quality –the ultimate goal ofthe high-κ research. Additionally, the EOT can be improved when the newhigh-κ oxide has an even larger κ value. Several high-κ oxides are currentlystudied such as, e.g. CeO2 [100], Pr2O3 [229], Y2O3 [230], Er2O3 [231],Lu2O3 [232], and LaAlO3 [233], the latter being one of the most promisingcandidates of the higher-κ dielectrics.

4.2 The (100)Si/LaAlO3 structure

Park and Ishiwara projected that the Si/LaAlO3 structure could be a pos-sible candidate [234] as inferred from depositing the first LaAlO3 thin films(κ=20-27) on (100)Si using molecular beam deposition (MBD) [235]. Later,Edge et al. concluded through an intensive study combining Auger electronspectroscopy, infrared absorption, MEIS, XPS, and HRTEM –an HRTEMimage is shown in Fig. 4.1– that as-grown LaAlO3 films deposited by MBDon (100)Si at a temperature of 100 ◦C have less than 0.2 A of SiO2 at theSi/LaAlO3 interface [233]. They also showed that these structures haveadditional favorable properties such as a 6.2 eV band gap of the LaAlO3

layer and band offsets of (1.8±0.2) eV for electrons and (3.2±0.1) eV forholes [121]. Lu et al. studied amorphous LaAlO3 films deposited on (100)Siby laser MBD in the temperature range 300-700 ◦C in oxygen containing

which indicates that there are no gross compositional varia-tions across the sample.

The MEIS spectrum of a 40-Å-thick LaAlO3 film onsilicon, deposited in the same way as the other amorphousLaAlO3 /Si samples, shows well separated lanthanum,aluminum/silicon, and oxygen backscatter peaks~Fig. 5!.When the lanthanum and oxygen peaks are plotted on thesame depth scale after corrections for cross section and con-centration~see the inset!, we see that the two elements havenearly identical depth distributions, indicating the near totalabsence of interfacial SiO2 . Quantitative modeling con-firmed this conclusion, to an upper limit of 2 Å of interfacialSiO2 . We note that the masses of aluminum and silicon aresimilar, making it difficult to discriminate between the twospecies in a buried layer using MEIS. Thus, although we canput tight constraints on the quantity of interfacial SiO2 , thereis a reduced sensitivity to interfacial LaSiOx at the interfacebetween the amorphous LaAlO3 and Si. The ability to avoidsubcutaneous oxidation provides an opportunity to controlla-bly introduce SiO2 at the interface and study its effect on theelectrical properties of the dielectric stack as a function ofthe interfacial SiO2 thickness.

L.F.E., D.G.S., C.H., G.L., Y.Y., and S.S. gratefully ac-knowledge the financial support of the Semiconductor Re-search Corporation~SRC! and SEMATECH through the

SRC/SEMATECH FEP Center. L.F.E. gratefully acknowl-edges an AMD/SRC fellowship. The portion of this workconducted at PNNL was carried out in the EnvironmentalMolecular Sciences Laboratory, a national scientific user fa-cility sponsored by the Department of Energy’s Office ofBiological and Environmental Research.

1D. A. Muller, T. Sorsch, S. Moccio, F. H. Baumann, K. Evans-Lutterodt,and G. Timp, Nature~London! 399, 758 ~1999!.

2C. A. Billman, P. H. Tan, K. J. Hubbard, and D. G. Schlom, inUltrathinSiO2 and High-K Materials for ULSI Gate Dielectrics, edited by H. R.Huff, C. A. Richter, M. L. Green, G. Lucovsky, and T. Hattori~MaterialsResearch Society, Warrendale, PA, 1999!, Vol. 567, pp. 409–414.

3A. I. Kingon, J.-P. Maria, and S. K. Streiffer, Nature~London! 406, 1032~2000!.

4D. G. Schlom and J. H. Haeni, MRS Bull.27, 198 ~2002!.5B.-E. Park and H. Ishiwara, Appl. Phys. Lett.79, 806 ~2001!.6B.-E. Park and H. Ishiwara, Appl. Phys. Lett.82, 1197~2003!.7X.-B. Lu, Z.-G. Liu, Y.-P. Wang, Y. Yang, X.-P. Wang, H.-W. Zhou, andB.-Y. Nguyen, J. Appl. Phys.94, 1229~2003!.

8L. F. Edge, D. G. Schlom, S. A. Chambers, E. Cicerrella, J. L. Freeouf, B.Hollander, and J. Schubert, Appl. Phys. Lett.84, 726 ~2004!.

9S. Stemmer and D. G. Schlom, inNano and Giga Challenges in Micro-electronics, edited by J. Greer, A. Korkin, and J. Labanowski~Elsevier,Amsterdam, 2003!, pp. 129–150.

10International Technology Roadmap for Semiconductors: 2003~Semicon-ductor Industry Association, San Jose, CA, 2003!.

11S. Guha, E. Cartier, M. A. Gribelyuk, N. A. Bojarczuk, and M. C. Copel,Appl. Phys. Lett.77, 2710~2000!.

12G. D. Wilk and R. M. Wallace, Appl. Phys. Lett.76, 112 ~2000!.13J. Kwo, M. Hong, A. R. Kortan, K. L. Queeney, Y. J. Chabal, R. L. Opila,

D. A. Muller, S. N. G. Chu, B. J. Sapjeta, T. S. Lay, J. P. Mannaerts, T.Boone, H. W. Krautter, J. J. Krajewski, A. M. Seregent, and J. M. Ro-samilia, J. Appl. Phys.89, 3920~2001!.

14S. Stemmer, D. O. Klenov, Z. Chen, D. Niu, R. W. Ashcraft, and G. N.Parsons, Appl. Phys. Lett.81, 712 ~2002!.

15W. Tsai, R. J. Carter, H. Nohira, M. Caymax, T. Conard, V. Cosnier, S.DeGendt, M. Heyns, J. Petry, O. Richard, W. Vandervorst, E. Young, C.Zhao, J. Maes, M. Tuominen, W. H. Schulte, E. Garfunkel, and T. Gustafs-son, Microelectron. Eng.65, 259 ~2003!.

16J. Kwo, M. Hong, B. Busch, D. A. Muller, Y. J. Chabal, A. R. Kortan, J.P. Mannaerts, B. Yang, P. Ye, H. Gossmann, A. M. Sergent, K. K. Ng, J.Bude, W. H. Schulte, E. Garfunkel, and T. Gustafsson, J. Cryst. Growth251, 645 ~2003!.

17M. Ritala, K. Kukli, A. Rahtu, P. I. Ra¨isanen, M. Leskela¨, T. Sajavaara,and J. Keinonen, Science288, 319 ~2000!.

18E. P. Gusev, M. Copel, E. Cartier, I. J. R. Baumvol, C. Krug, and M. A.Gribelyuk, Appl. Phys. Lett.76, 176 ~2000!.

19S. Guha, E. Cartier, N. A. Bojarczuk, J. Bruley, L. Gignac, and J. Kara-sinski, J. Appl. Phys.90, 512 ~2001!.

20S. J. Wang, C. K. Ong, S. Y. Xu, P. Chen, W. C. Tjiu, J. W. Chai, A. C. H.Huan, W. J. Yoo, J. S. Lim, W. Feng, and W. K. Choi, Appl. Phys. Lett.78,1604 ~2001!.

21S. J. Wang and C. K. Ong, Appl. Phys. Lett.80, 2541~2002!.22H. Li, X. Hu, Y. Wei, Z. Yu, X. Zhang, R. Droopad, A. A. Demkov, J.

Edwards, K. Moore, W. Ooms, J. Kulik, and P. Fejes, J. Appl. Phys.93,4521 ~2003!.

23J. Y. Dai, P. F. Lee, K. H. Wong, H. L. W. Chen, and C. L. Choy, J. Appl.Phys.94, 912 ~2003!.

24L. Yan, H. B. Lu, G. T. Tan, F. Chen, Y. L. Zhou, G. Z. Yang, W. Liu, andZ. H. Chen, Appl. Phys. A: Mater. Sci. Process.77, 721 ~2003!.

25L. F. Edge and D. G. Schlom~unpublished!.26The nominal film thicknesses given were calculated from the fluxes of the

molecular beams~measured by a quartz crystal microbalance! assumingthe amorphous LaAlO3 films had the density of crystalline LaAlO3 . Theareal density~atoms/cm2! of lanthanum and aluminum in the films wasconfirmed by RBS. The thicknesses of the amorphous LaAlO3 films werealso measured by x-ray reflectivity and show 1.5–2 times the thicknessesdetermined assuming the density of crystalline LaAlO3 . These differenceswill be discussed elsewhere.

27J. Lettieri, J. H. Haeni, and D. G. Schlom, J. Vac. Sci. Technol. A20, 1332~2002!.

28B. B. Stefanov, A. G. Gurevich, M. K. Weldon, K. Raghavachari, and Y. J.Chabal, Phys. Rev. Lett.81, 3908~1998!.

FIG. 4. Cross-sectional HRTEM image of 10 Å of amorphous LaAlO3 on~001! Si showing no interfacial layer between the LaAlO3 film and silicon.

FIG. 5. MEIS energy spectrum of 40-Å-thick amorphous LaAlO3 on ~001!Si along with the model calculated assuming an SiO2-free interface. Insetshows lanthanum and oxygen depth distributions.

4631Appl. Phys. Lett., Vol. 84, No. 23, 7 June 2004 Edge et al.

Downloaded 11 Jan 2007 to 134.58.253.113. Redistribution subject to AIP license or copyright, see http://apl.aip.org/apl/copyright.jsp

Figure 4.1: Cross-sectional HRTEM image of 10 A of amorphous LaAlO3 on(100)Si showing no interfacial SiOx layer. The figure is taken from Ref. [233].

4.2 The (100)Si/LaAlO3 structure 141

(a) (b)

Figure 4.2: Cross-sectional HRTEM pictures of LaAlO3 films deposited on (100)Siat different substrate temperatures: (a) 500 ◦C and (b) 700 ◦C. The figures aretaken from Ref. [236].

ambient. Here, HRTEM measurements revealed the presence of an interfa-cial layer, with thickness increasing with substrate temperature, introducedduring film deposition [236], as shown in Fig. 4.2. The LaAlO3 films re-mained amorphous after post-deposition annealing (PDA) at 1000 ◦C inN2 or O2, but were found to exhibit a better thermal stability in contactwith Si when annealed in an N2 ambient. Applying still a different de-position method, another work [237] reported on a study of LaAlO3 filmsgrown on Si using CVD. Also here, the analysis revealed the occurrenceof an interlayer between the oxide and the Si substrate inferred as madeup of compositionally graded La-Al-Si-O silicate rather than pure SiOx.In a recent paper Sivasubramani et al. reported on a study of amorphousLaAlO3 grown on (100)Si by MBD studied as a function of post depositionrapid thermal annealing (RTA) for 20 s in flowing N2 [238]. The LaAlO3

films were capped in situ with ∼100 A of Al2O3 in order to protect thefilm surface from hydroxyls. They found the (100)Si/LaAlO3 system to bevery stable. A change in the structure of LaAlO3 from amorphous to poly-crystalline takes place only after a 935 ◦C RTA, nicely illustrated by theAFM images and XRD spectra shown in Fig. 4.3. Through SIMS profilingas a function of thermal treatment, shown in Fig. 4.4, they showed thatupon annealing at higher Tan, the crystallization is followed by La and Alpenetration into the Si after RTA at temperatures 950 ◦C.

In light of the above results, it appears of interest to get more in depthinformation on the true interfacial nature of Si/LaAlO3 structures and the


Figure 4.3: (a) AFM images (area=25 µm2) of the capped Al2O3/LaAlO3/Si(100)stack before and after thermal treatments at different RTA conditions. (b) XRDspectra of the stack following RTA treatments in flowing N2. The peaks correspondto all of the expected reflections of crystalline LaAlO3. The figures are taken fromRef. [238].

incidence angle of 60°. A positive secondary ion count forLa, Al, and Si was monitored as a function of depth. A stylusprofilometer was utilized to determine the depth scale as afunction of sputtering time. La and Al concentrations weredetermined from comparison with ion implanted standards.14

Figure 1sad shows the surface morphology evolution ofthe Al2O3/LaAlO3/Si s001d stack evaluated by AFM as afunction of the RTA conditions. After a 900 °C, 20 s, RTA inN2, a change of the surface topography and a root-mean-squaresrmsd roughnesssRrmsd of 4 Å is observed. At orabove 1000 °C, 10 s RTA, a grain size of,0.5–3mm diam-eter, andRrms,40 Å is observed.16

The XRD 2u scans of the stack before and after RTA inN2 are shown in Fig. 1sbd. Above 900 °C RTA, XRD peakscorresponding to polycrystalline LaAlO3 are observed. Theseresults are consistent with previous reports from LaAlO3films produced by atomic layer deposition,17 pulsed laserdeposition,18 and chemical vapor deposition.19 Since theAl2O3 capping layer is relatively thin compared to theLaAlO3 film, no crystalline reflections of the capping layerare detected in the 2u range analyzed.16

Figure 2 shows the back side SIMS concentration versusdepth profile for Al2O3/LaAlO3/Si s001d stack before andafter RTA in flowing N2. The 1000 Å thick LaAlO3 film

essentially provides an infinite source of La and Al for dif-fusion into bulk Si.

Figure 2sad shows the Al concentration in the siliconsubstrate as a function of depth at different annealing tem-peratures. Up to,935°C, 20 s RTA in N2, the penetration ofAl into the Si substrate bulk is below detectible limits. How-ever, after a 950 °C, 20 s, RTA in N2, substantial Al penetra-tion is observed. The concentration of Al is found to be.1016 atoms/cm3 up to a depth of,1000 Å from the Si/LaAlO3 interface into the bulk Sis001d substrate after a1000 °C, 10 s N2 anneal, and would therefore be a concernfor mobility degradation.8 The penetration depth of Al is,1500 Å for the 1025 °C, 20 s, RTA. These results are gen-erally consistent with previously reported Al interdiffusionfrom similar RTA treatments of Al2O3 films in contact withSi, and may lead to mobility degradation.10 Additionally, ourresults indicate that the penetration depth is,400 Å deeperbefore the Al concentration in bulk silicon is reduced to,1017 atoms/cc.

Figure 2sbd shows the corresponding La profile wheresignificant penetration is observed at RTA temperaturesù950°C. The penetration of La was,400 Å for thehighest-temperature RTA condition examined here of1025 °C, 20 s. These results are consistent with the relativemasses of the diffusing species,viz. Al diffusing deeper intothe Si substrate compared to La, as well as recent reports onthe silicidation of La upon thermal treatments.20–22 We alsonote that Si diffusion into the LaAlO3 cannot be ruled out in

FIG. 1. sad AFM imagessarea=5mm35 mmd of the Al2O3/LaAlO3/p-Sis001d stack before and after thermal treatments at different RTA conditions.sbd XRD spectra of the stack following RTA treatments in flowing N2. Thepeaks correspond to all of the expected reflections of crystalline LaAlO3 andare indexed with pseudocubic indices.

FIG. 2. Backside SIMS concentration vs depth profile forsad Al and sbd Labefore and after RTA treatments of 100 Å Al2O3/1000 Å LaAlO3/p-Si s001dstack in flowing N2. The depth scale up to,0.4 mm corresponds to the Sisubstrate, followed by the 1000 Å thick LaAlO3 film. SIMS results are forthe same films described in Fig. 1.

201901-2 Sivasubramani et al. Appl. Phys. Lett. 86, 201901 ~2005!

Downloaded 11 Jan 2007 to 134.58.253.113. Redistribution subject to AIP license or copyright, see http://apl.aip.org/apl/copyright.jsp

Figure 4.4: Backside SIMS concentration vs. depth profile for (a) Al and (b) Labefore and after RTA treatments of Al2O3/LaAlO3/Si stacks in flowing N2. Thedepth scale up to ∼0.4µm corresponds to the Si subtrate, followed by the 1000 Athick LaAlO3 film. The figures are taken from Ref. [238].


consequences of PDA treatments. The scope of the present work is to attainstructural information on atomic scale by monitoring paramagnetic pointdefects in (100)Si/LaAlO3 structures with the oxide thin films grown byMBD as a function of the PDA temperature (Tan). We will demonstrate,through the occurrence/absence of interface specific (Pb0, Pb1) and/or in-terlayer associated (EX) point defects, the appearance, additional growthand modification of a SiOx interlayer at Tan>800 ◦C. At anneal temper-atures up to 800 ◦C, the Si/LaAlO3 interface is found to be abrupt andthermally stable. A SiOx interlayer is formed after annealing in the tem-perature range Tan∼800-860 ◦C. However this interlayer starts to breakdown after PDA at Tan∼930 ◦C, possibly resulting in silicate formation(Tan∼1000 ◦C).


4.3.1 Samples

Details about the samples studied can be found elsewhere [233, 121]. Inshort, uniform LaAlO3 thin films (10-40 nm) were grown by MBD in anEPI 930 MBE chamber modified for the growth of oxides [239] on Si sub-strates. The latter were one-side polished commercial 8 inch Si wafers(p-type; na∼(3-6)×1015Bcm−3). Prior to deposition the native SiO2 onthe Si wafers was in situ thermally removed in ultrahigh vacuum at a sub-strate temperature of 950 ◦C (as measured with a pyrometer). Subse-quently, using elemental sources, La, Al, and molecular oxygen (99.994%)at a background pressure 6×10−8 Torr were codeposited on the substrateat ∼100 ◦C. The La and Al fluxes from the effusion cells were each 2×1013

at.cm−2s−1. As analyzed by Rutherford backscattering spectroscopy, thisresulted in closely stoichiometric layers (La:Al mol ratio=1±0.05).


From these wafers, ESR slices of 2×9 mm2 main area were cut with their 9-mm edge along a <011> direction. Cutting damage was removed throughselective chemical etching of backside and edges. Thermal stability of de-posited LaAlO3 layers and interfaces was analyzed by subjecting samplesto isochronal (∼10 min) PDA at desired temperatures between 630 and1000 ◦C generally in a 1 atm N2 + 5% O2 (99.995%) ambient or pureN2 (99.999%) using a conventional resistively heated laboratory facility.Several sets of samples were used for the various thermal steps. As an


additional test related to potential ESR activation/maximalization of dia-magnetic defects some samples were subjected for ∼10 s to 10.02 eV VUVphotons (flux ∼5×1014 cm−2s−1) obtained from a Kr resonant dischargelamp or to short positive corona charging (3 µA for 10 s) in room ambient.

Conventional cw slow-passage K-band ESR measurements were carriedout at 4.2 K, as described elsewhere [29], for the applied magnetic field ~Brotating in the (011)Si substrate plane over an angular range ΦB=0-90◦,with respect to the [100] interface normal ~n. A co-mounted Si:P referencesample of g(4.2 K)=1.99869±0.00002 was used for g factor and spin densitycalibration, with the latter performed through orthodox double numericalintegration of the detected derivative-absorption spectra. The attained ab-solute and relative accuracy is estimated at ∼30% and ∼10%, respectively.Typically, an ESR sample was comprised of 10-12 slices HF (5% in H2O)dipped immediately before taking ESR data. Signal averaging (∼typical100 scans) was routinely applied to enhance spectral quality.

4.4 Experimental results and analysis

4.4.1 Observed defects: Pb’s and EX

Figure 4.5 presents an overview of representative ESR spectra, observedwith ~B‖~n on the as-deposited (100)Si/LaAlO3 structures and after differ-ent PDA steps in N2 + 5% O2 ambient. Within spectral accuracy, no ESRactive defects could be observed on the as-grown samples, the situationremaining unaltered for annealing up to Tan≤800 ◦C. However, upon an-nealing in the temperature range Tan∼860-970 ◦C, a resonance signal isobserved at zero crossing g=2.0060±0.0001 with ∆Bpp=(7±1) G, exhibit-ing distinct angular anisotropy. To trace the signal’s origin, a coarse g mapwas composed for ~B rotating in the (011) plane. As shown in Fig. 4.6,the obtained data could be well fitted by an axial symmetric system (cf.solid curves in Fig. 4.6) with principal g values g‖=2.0017±0.0001 andg⊥=2.0082±0.0001. These results match those obtained for the g patternof the Pb0 defect in standard thin Si/SiO2 systems, as elaborated on insection 1.2.1 (see Fig. 1.6), leaving little doubt about the signal’s origin.It appears that the (100)Si/LaAlO3 interface has become Si/SiO2-type,evidencing that a SiOx-type interlayer has formed.

Here, we should add that some of the PDA steps at representative tem-peratures were also carried out in pure (99.999%) N2. Generally, these ledto similar ESR results as those for the O2/N2 ambient, with no outspokeneffect on Pb0 appearance. So, while it cannot entirely be excluded that the

4.4 Results and analysis 145

25

7290 7300 7310 7320 7330 7340 7350 7360 7370

1000 °C

(100)Si/LaAlO3 20.4 GHz; 4.2 K

Si:P

970 °C

940 °C

908 °C

Tan=860 °C

as-grown

2.0025

2.0060Pb0

g=1.99869

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

EX

Fig. 1.

Figure 4.5: Representative derivative-absorption K-band (Bm=0.6 G; Pµ∼2.5nW ) ESR spectra measured at 4.2 K with applied magnetic field ~B perpendicu-lar to the interface of (100)Si/LaAlO3 structures subjected to different steps ofpost-deposition annealing in N2 + 5% O2 (10 min). Spectral heights have beennormalized to equal marker intensity and sample area. The signal at g=1.99869stems from a co-mounted Si:P marker sample.


26

Fig. 2.

0 20 40 60 802.001

2.002

2.003

2.004

2.005

2.006

2.007

2.008

2.009

2.010

2

1

1

[211][111][211][100]

g va

lue


Figure 4.6: Coarse angular g map of the observed resonances, ascribed to Pb0for ~B rotating in the plane with respect to the interface normal ~n observed in a(100)Si/LaAlO3 structure subjected to a post-deposition anneal at ∼900 ◦C for∼10 min. The curves represent the optimized fitting of the various branches foraxial symmetry yielding g‖=2.0017±0.0001 and g⊥=2.0082±0.0001, within ex-perimental accuracy in agreement with the Pb0 data for standard thin thermaloxide/Si structures, affirming the Pb0 assignment. The added numbers indicaterelative branch intensities expected from the standard defect model, also matchingexperimental observation.

presence of O2 in the anneal ambient might have promoted the Si/SiOx

interface formation, its effect is not conclusive.As can be seen from Fig. 4.5, one more signal is observed after annealing

in the range Tan∼888-940 ◦C at g=2.0025±0.0001. It could be conclusivelyidentified as the EX signal from the observed accompanying hf doublet(∼16.1 G splitting) centered around this g value, typical for this defect.This is evident from Fig. 4.7 showing a zoomed-in representative spectrumof the EX center observed in the sample subjected to a 940 ◦C anneal andadditional VUV irradiation. As outlined in section 1.2.2, the EX defectis an SiO2 associated center, well known from studies of bulk SiO2 andvarious kinds of Si/SiO2 structures [77, 67, 68, 69, 71, 36, 63, 72, 73, 75].

In the as-deposited and PDA treated samples, no LaAlO3-specific pointdefects could be observed. Quite surely, though, point defect sites will bepresent, but reside in a diamagnetic state, making them invisible for ESR


27

7270 7280 7290 7300 7310 7320

16.1 G

g=1.99869

Si:PEX

Pb0

20.4 GHz; 4.2 K940 °C anneal + VUV

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

Fig. 3.

Figure 4.7: Derivative-absorption K-band ESR spectrum improved through sig-nal averaging (∼200 scans) of the observed defects measured at 4.2 K on(100)Si/LaAlO3 subjected to post deposition annealing in N2 + 5% O2 (10 min)at 940 ◦C and additional VUV irradiation.

detection. So to probe deeper it was aimed to maximally reveal defects inthe oxide as well as at the interface, through subjecting some samples toadditional VUV irradiation –known to be a most efficient means for bothoxide [240, 168, 241] and interface defects in Si/SiO2– to photo-dissociateH-terminated dangling bonds and to possibly additionally unveil weak orstrained bonds [169, 242]. Still any evidence for LaAlO3-associated defectsremained lacking as no ESR spectral changes were noticed.

Importantly as one more result, this finding thus also indicates that noPb-type defect system remained hidden or was left partially passivated –ascould be expected from the H-free fabrication process of the samples. In thesame spirit, some samples where additionally subjected to positive coronacharging, applied for a very short time though to avoid impairing H-relatedeffects inherent to such method [117], to possibly put diamagnetic defectsin the LaAlO3 in the correct charge state for ESR detection. Yet, in spiteof these efforts, detection of any LaAlO3-related defects still failed.


4.4.2 Defect densities vs. Tan

Monitoring the densities of the observed defects, Pb0 and EX, as a functionof PDA treatment revealed various interesting aspects about the nature andstability of the (100)Si/LaAlO3 interface. The main PDA results (5% O2

+ N2) are compiled in Fig. 4.8 showing the density of observed defects asa function of PDA temperature. Various items are worth noting:

No ESR signals, in particular no Pb-type defects could be observed in theas-deposited Si/LaAlO3 structure. On the basis of the latter criterion it ev-idences, with atomic level sensitivity, that there is no SiOx-type interlayerpresent or that this interlayer is at least substantially thinner than [243]3 A –an abrupt interface– which is in good agreement with previous re-sults [233]. This is in sharp contrast with other stacks of (100)Si withhigh-κ layers, such as Al2O3, HfO2, and ZrO2, where the presence of suchan interlayer appeared inevitable, even in the as-deposited state, which wasrevealed by ESR [112, 113, 118, 77, 115, 114, 116]. The abrupt interfaceremains unaltered even during subsequent annealing up to Tan∼800 ◦C,indicating that the interface is thermally stable.

After annealing at Tan∼860 ◦C an SiOx-type interlayer has formed, asevidenced by the observation of Pb0 defects. Upon annealing at a slightlyhigher temperature Tan∼888 ◦C, the SiO2-associated EX defect appears,so, delayed over ∼30 ◦C in terms of Tan. This observation corroborates thepresence of an SiOx-type interlayer and also indicates an additional growthof the interlayer. The defect was not observed in the sample annealed atTan∼860 ◦C even though the presence of an SiOx-type interlayer in thissample is signaled by the observation of Pb0 defects. It suggests that aminimal thickness of the SiOx interlayer is needed for (ESR) detectableformation of EX defects, at least more substantial than required for effec-tive Si/SiOx-interface formation. The need of a minimal thickness of SiO2

for EX detection has been reported before for dry thermal SiO2 on Si [69].Thus the retardation in EX appearance vis-a-vis Pb0 would indicate anadditional growth (or modification) of the interlayer. The interlayer thick-ness, however, is unknown. In broader context, it is interesting to note thatthe generation of EX centers upon annealing at elevated temperatures inoxygen-containing ambients appears symptomatic for stacks of high-κmetaloxide layers on Si [77].


20 600 800 1000

0

2

4

6

8

10

12

14

10 min anneal (5% O2 in N2) Pb0

EX

defe

ct d

ensi

ty (1

011 c

m-2)

Tan (°C)

Figure 4.8: Compilation of the inferred defect densities in (100)Si/LaAlO3 enti-ties as a function of the post deposition isochronal annealing in N2 + 5% O2 (10min) for Pb0 and EX centers, represented by black closed and red open symbols,respectively. The solid and dashed traces are Gaussian curves, merely meant toguide the eye in exposing the peaking in defect generation and the somewhat lag-ging behind (∼30 ◦C) of EX production vis-a-vis Pb0 appearance. Data points atzero defect density symbolize failure of signal detection.


4.5 Discussion

4.5.1 Interlayer formation and disintegration

Monitoring of occurring point defects as a function of PDA treatment,additional VUV irradiation, and corona charging revealed interesting infor-mation on the annealing induced structural/compositional changes in theinterface region. Compiling the thermal evolution of the observed defectdensities, hereby displaying all main results, Fig. 4.8 will guide the discus-sion and interpretation. Prominently, the defect density vs. Tan curvesshow a peaked behavior, simulated by Gaussian curves. The maximumPb0 density, [Pb0]∼(1.3±0.3)×1012 cm−2, is obtained at Tan=900 ◦C whilethe maximum EX density, [EX]∼(1.0±0.3)×1012 cm−2, is obtained at aslightly higher PDA temperature Tan=940 ◦C. On the account of Pb0, itappears that for annealing from Tan>900 ◦C onward, the interface startsto break up first, where it is interesting to note that, in the range Tan∼900-940 ◦C, the EX density still increases while the Pb0 density already startsto decrease.

So, the overall picture emerging from the curves in Fig. 4.8 is that, ascompared to Pb0, the manifestation of the EX peak is delayed in termsof Tan. The true character of the Si/SiOx-type interface is disrupted first(elimination of interfacial Si dangling bonds), but for Tan≥940 ◦C, thedefects rapidly disappear altogether, pointing to drastic disintegration ofthe interfacial region, i.e., elimination of the ’pure’ SiOx component. Forclarity, this disappearance of the ESR-active centers is not due to inadver-tent H-passivation, as verified by applying additional VUV irradiation aftersome PDA steps. The structural/chemical change breaking up the inter-layer seems to first affect the Pb0 defects located at the interface before itaffects the interlayer associated EX defects.

4.5.2 Influence of oxide crystallization and interdiffusion

It is interesting to put the currently acquired atomic level information interms of occurring point defects in the perspective of previous morpho-logical/compositional studies [233, 238, 235, 236] on the Si/LaAlO3 struc-ture. Using such methods, recent work [238] has investigated the thermalstability of (100)Si/LaAlO3/Al2O3 stacks against RTA in N2 ambient at850-1040 ◦C for 10-20 s. As observed by atomic force microscopy and X-ray diffraction (Fig. 4.3), the work reports that RTA from 900 ◦C onwardstarts to induce changes in surface morphology, together with initiation of

4.5 Discussion 151

the transformation of the amorphous to polycrystalline state, in agreementwith previous results on LaAlO3 films produced by different methods. SIMSprofiling shows that substantial penetration of Al and La atoms into the Sisubstrate occurs for RTA at or above 950 ◦C, with in fact any penetrationeffects remaining below the detection limit for Tan≤935 ◦C, as can be seenfrom Fig. 4.4.

Our ESR data may well fit within this morphological picture: In termsof Tan, we may link the initiation of the formation of a Si/SiOx-type in-terface with the very early onset of LaAlO3 film crystallization, followedby some more substantial SiOx-type interlayer growth with increasing Tan.For clarity, though, the appearance of the Pb0 centers cannot be directlylinked to the grain boundary regions per se, as this would conflict with theobserved ESR spectral anisotropy in registry with the crystalline (100)Sisurface. Then, for Tan further increasing above ∼940 ◦C, the progressinginterdiffusion chemically destroys the pristine nature of the SiOx compo-nent (possibly silicate formation) resulting in the obliteration of the SiO2(x)-specific point defects. It is possible that the onset temperature for La andAl out diffusion in the current case is somewhat lower considering the ap-plied longer PDA treatment times (10 min) as compared to previous RTAwork (10-20 s).

Also, within the interpretation, it appears we detect the initiation ofcrystallization of the amorphous LaAlO3 film somewhat at lower Tan thanin previous work [238]. Again, this may partly have resulted from theapplied longer anneal times in conjunction with the presence of O2 in theanneal ambient. Yet, it may as well bear out the fact that ESR is proneto detect interfacial reshaping in a very embryotic state, ahead of standardmorphological/compositional methods.

4.5.3 Absence of Pb1 defects

The observation of the Pb0 defect indicates that a Si/SiO2-type interfacehas formed. As demonstrated, the obtained ESR parameters such as gvalue, line width and defect density are indeed very similar to those char-acteristic for thermal Si/SiO2 interfaces. There is, however, one remarkableapparent dissonance: No Pb1 (g1=2.00577, g2=2.00735, g3=2.0022) centerscould be detected. As outlined in section 1.2.1, the Pb1 center also con-cerns an unpaired sp3 hybrid at a threefold Si atom part of a strained Si–Sidimer (≡Si–Si•=Si2), thus basically chemically identical to the Pb0 center,yet physically different, e.g., regarding hybrid orientation, bond strain, andstructural relaxation [38]. Generally, the Pb0 and Pb1 centers are almost


invariably observed in tandem at the (100)Si/SiO2 interface, although rela-tive intensities may vary. So, the question arises as to the absence of the Pb1defect, at least below ESR sensitivity, at the current Si/SiO2(x)-type inter-face formed during PDA of (100)Si/LaAlO3 entities. There may be severalreasons. One considered here concerns interfacial strain. Pertinently, thePb1 center may be less prominent, even absent beyond the detection limitin Si/SiO2 entities grown at low oxidation temperatures or after partic-ular PDAs. In previous work [244], it has been outlined this would notjust be the result of the small oxide thickness per se, as compared to morestandard SiO2 thicknesses (≥10 nm). Rather, Pb1 formation would requirea minimum amount of oxide (interface) relaxation as generally inherentlyestablished during oxidation at high temperatures. So, as a possibility, suf-ficient interface relieve may not have been attained in the current case, noteven for PDA steps up to 950 ◦C. Possibly, in this the top LaAlO3 layerplays a constraining role.

4.6 Conclusions

The ESR technique has been successfully applied to assess the nature ofthe interface in (100)Si/LaAlO3 entities and thermally induced alterations.It is found that the (100)Si/LaAlO3 structure exhibits a high quality androbust interface in terms of dangling bond-type interface defects, stableunder extended thermal anneals up to ∼850 ◦C, even in O2-containingambient.

Upon annealing at Tan≥860 ◦C a SiOx-type interlayer is formed as ev-idenced by the appearance of Pb0 interface defects, an effect likely fore-running or heralding the onset of crystallization of the a-LaAlO3 film. Atsomewhat higher anneal temperature, the EX defect is observed the de-lay in appearance vis-a-vis Pb0 pointing to additional growth of the SiOx-type interlayer. However upon annealing at temperatures Tan≥930 ◦C,the Si/SiOx-nature of the interlayer starts to break up again, resulting infast, likely compositional, transformation with increasing Tan as signalledby the disappearance of the SiOx related defects altogether. It is ascribedto invasive diffusion of La and Al into the Si substrate, possibly resultingin silicate (interlayer) formation.

Generally, ESR has emerged as a viable technique to trace thermally in-duced structural/compositional alterations in Si/insulator entities throughprobing the incorporation of interface/interlayer related point defects. Themethod takes a separate position in the sense that it may reveal such

4.6 Conclusions 153

changes on atomic level where standard morphological/structural analyzingtools simply fall short, or detect these in an earlier stage before the lattermay achieve.

Chapter 5

Origin of optical absorptionin the range 4.8-4.9 eV and5.7-5.9 eV in silica implantedwith Si or O.

5.1 Introduction

The optical activity of defects in SiO2 has been of interest for many yearsdue to the crucial role they may play in the performance of SiO2-based opto-electronic devices [245, 246, 247]. Consequently, the optical absorption ofthese glasses has been intensively studied. An extensive overview is given inRef. [248]. The optical spectra, however, generally consist of broad bands,rather than of separate sharp lines. Because of this, the disentanglement ofthe various optical absorption bands and their attribution to certain defectcenters is a difficult task. Quite often optical absorption measurements arecombined with ESR in an attempt to correlate optical absorption bandswith ESR spectra. It should be noted, however, that ESR spectroscopyonly detects paramagnetic defect sites, while this is not a restriction foroptical absorption.

Here we will compare the optical absorption spectra in the range 4.8-4.9 eV and 5.7-5.9 eV , measured and analyzed by Magruder et al., withthe ESR spectra of two series of silica samples implanted with variousconcentrations of Si or O.

The broad absorption band detected around 4.8 eV was first attributedto NBOHCs [249]. But, while a good correlation was reported between

155

156 Chapter 5: origins of optical absorption in Si or O implanted silica

this optical absorption band and the ESR signal of the NBOHCs in somestudies, a poor correlation was found in others [250]. Based on combinedoptical and ESR measurements Hosono and Weeks attributed this band toPORs [251]. Alternatively it was suggested that the 4.8 eV band is causedby interstitial ozone molecules, as the absorption spectrum of gas-phase O3

was found to be similar [252, 253]. Skuja et al. suggested that the 4.8 eVband consists of two closely overlapping bands attributed to NBOHCs andinterstitial O3 [254]. They also reported [255] interstitial oxygen molecules,O2, in silica and Awazu and Kawazoe [252] describe the interaction of theinterstitial molecules with the interstitial oxygen atom to form the ozonemolecule. Thus there are at least three optical absorption bands in therange 4.8-4.9 eV attributed to O related defects including NBOHCs, PORs,and interstitial O3.

The optical absorption band between 5.7 and 5.9 eV has often beenassigned to E′ type defect centers. Weeks and Sonders [256] demonstratedthat the amplitude of the first derivative of the ESR absorption spectrumof the E′ center and the intensity of an optical absorption band at 5.85eV had a correlation coefficient R=0.999. They showed that the energyof the band maximum increased from 5.80 to 5.85 eV and the full widthat half maximum (FWHM) amplitude decreased from ∼0.9 to ∼0.63 afterbleaching with a mercury lamp. The authors [256] further suggested thatthere were three or more optical absorption bands between 5.5 and 6.0 eV .Based on bleaching experiments with 5.0 eV laser light on type III andIV silica samples, Weeks et al. [257] proposed a band with a maximumbetween 5.8 and 5.9 eV that was not paramagnetic. One work reports ona shift in optical band maximum with changes in the E′γ features (zerocrossing g value and line shape), which they attributed to the modificationof the E′γ structure [258]. The energy of the band maximum increasedfrom 5.75 to 5.83 eV . Nishikawa et al. [259], using type III and IV silicasamples, observed that the absorption maximum for the E′ optical bandwas ∼5.7 eV . Their data on the correlation of the density of the E′ centerand the band maximum amplitude gave a correlation coefficient R∼1 forboth types of silica. The absence of experiments on the effect of bleachingand anneal treatments on the band maximum for their samples leaves openthe question of the contribution of other bands between 5.4 and 6.0 eV tothe energy of the band maximum.

Magruder and his collaborators [260, 261, 262] have measured the op-tical absorption in type III silica samples in which the excess Si, x, was0.02 at.%<x<3 at.% and excess oxygen, y, was 0.02 at.%<y<3 at.% in

5.1 Introduction 157

0.1 10

5

10

15

20

25

30

Si O

Abs

orpt

ion

coef

ficie

nt (1

02 cm

-1)

log(at.%)

4.83 eV band

Figure 5.1: The amplitudes for the band fit data for a 4.83 eV band as a functionof the logarithm of the atomic percent of implanted ions for both Si and O implantedsilica samples

0.1 10

20

40

60

80

100

log(at.%)

Abs

orpt

ion

coef

ficie

nt (1

02 cm

-1)

5.85 eV band

O Si

Figure 5.2: The amplitudes for the band fit data for a 5.85 eV band as a functionof the logarithm of the atomic percent of implanted ions for both Si and O implantedsilica samples


the layer in which the implanted ions were located. They then used theabsorption bands reported in literature, attributed to intrinsic states, tofit the observed absorption in a series of samples implanted with O and aseries of samples implanted with Si. Here the word intrinsic means thatthe states are a consequence of different configurations of Si and O. Linearand non-linear fitting techniques were used. From these fits the absorptioncoefficient for the 5.85 and 4.83 eV bands were inferred.

Here we will compare variations in the optical absorption coefficients asa function of implanted ion concentration, shown in Figs. 5.1 and 5.2, withthe variations of defect densities of occurring point defects detected by ESRin samples cut from the optical samples. All information about the opticalabsorption measurements and the analysis of the optical data can be foundin Refs. [260, 261, 262, 263, 264].


5.2.1 Samples and ion implantation

Multiple energy implants of Si and O were employed to produce an approx-imately constant volume concentration of the implanted element into theface of 2-cm diameter disks of about 1 mm thick type III (Corning 7940)silica. The computer code, PROFILE, was used to calculate the energiesand concentrations required to form this constant concentration. The ionenergies varied from 35 to 320 eV , and the resulting depth profiles wereconstant to within ±5% over a depth range 60 to 640 nm below the SiO2

surface for the Si implants and from 100 to 670 nm for the O implants. A 2cm diameter face of each sample (Corning 7940) was implanted uniformly.Implantation parameters and nominal area densities as a function of energyhave been previously reported [260, 262] as well as implant conditions. Herewe give the at.% concentration in the implanted layer in Table 5.1.


Conventional cw absorption-derivative K (∼20.3 GHz) and Q-band (∼33GHz) ESR measurements were performed in the temperature range 77-100 K. The modulation amplitude Bm and incident microwave power Pµwere reduced to levels at which neither signal distortion nor saturation wasobserved. Since the accurate determination of defect spin densities is a mainaim in the current work, the latter is very important when point defects


Table 5.1: Concentration in at.% of Si or O uniformly implanted in a surfacelayer of SiO2 of ∼ 600 nm.

Implanted Si Implanted O(at.%) (at.%)0.025 0.0350.05 0.070.25 0.350.5 0.71.5 2.1

with spin relaxation times >10−4 s such as the E′ center in silica [265] areconcerned.

The ESR samples were bars of about 1×1×16 mm3, cut from the centersof the 2 cm diameter optical samples. For g factor and defect (spin) densitycalibration, use was made of a calibrated Si:P marker sample co-mountedwith the sample and recorded in the same trace. This arrangement makesthe determination of the relative defect densities quite reliable. Signalaveraging was applied (typically 50-100 scans for K-band and 100-300 scansfor Q-band) to increase the signal to noise ratio.

As stated, a key result expected from ESR spectroscopy is a reliableand accurate inference of the densities of the various types of implantationinduced spin active defect centers. Generally these various ESR compo-nents overlap and, as well known from ESR spectroscopy, in fact from anyspectroscopy, their reliable decomposition poses a formidable task for whichthere is no straightforward remedy. It is very rare that the observed ESRspectrum would be comprised of only a few, in magnetic field, sufficientlyseparated individual components to enable each of them to be reliably in-tegrated separately. Clearly, we cannot handle the problem by just tryingto fit the experimental spectrum by a sum of a number, n, of theoreticalspectra through a statistically optimizing program. A satisfactory fit canalways be obtained, no wonder, in view of the many adjustable parametersinvolved (≥4 for each component, a priori all unknown), but must be re-garded as unreliable, to say the least. Instead, in good practice, we resortto experimental help, i.e., measuring as a function of some varied externalparameters such as microwave frequency, temperature, and Pµ, to help in


discriminating the various responses and to get a good impression of theirseparate appearance. In addition, computer simulation of some ’known’components –powder patterns in the current case– may help. Generally,in this work with respect to defect density inference, the best results areobtained when determining the densities of separate signals relative to theSi:P intensity standard through orthodox double numerical integration ofthe measured dPµ/dB spectra. (Obviously, one cannot solely rely on thecomponent amplitude.) Since this was often not possible due to componentoverlap, most spectra had to be computer simulated first, thus rendering itmore difficult to get accurate defect densities.

5.3 ESR results

5.3.1 Oxygen implanted samples

In the oxygen implanted samples the E′γ center [266], the POR [267, 81],the NBOHC [82], and an unknown signal (OS) were observed. Some of theK and Q-band spectra recorded for the different oxygen implanted samples

2.1 at.%

0.07 at.%

0.35 at.%

0.07 at.%

0.035 at.%

g=1.99889

Si:PO implanted20.5 GHz; 77 K

10 G

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

E'

2.1 at.%

0.7 at.%

0.35 at.%

0.07 at.%

0.035 at.%

O implanted~33 GHz; 100 K

g=1.99891

Si:P

20 G

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

Figure 5.3: Overview of measured K and Q-band ESR spectra on fused silicafor the different O implantation doses. As a measure of dose, the curves arelabeled by the calculated at.% of O atoms implanted in a layer ∼600 nm thick,over which the concentration is approximately constant. The size of all samplesstudied was nominally identical. The signal at g=1.99891 stems from a co-mountedSi:P marker sample. Signals are scaled to equal marker intensity.

5.3 Results 161

are shown in Fig. 5.3.Both the K and Q-band spectra of the OS exhibit powder pattern prop-

erties (cf. Fig. 5.4), and both could be readily and consistently simulatedby an effective spin S=1/2 center with one set of g matrix principal valuesgiven as g1=2.0022, g2=2.0033, and g3=2.0088. Examples of the signalsimulation are shown in Fig. 5.4. The center could only be observed inthe samples implanted with the larger oxygen concentrations (≥0.35 at.%),from where we suggest that the center is oxygen excess related.

As mentioned before, due to component overlap, the NBOHC, POR,and OS had to be simulated to determine the proper spin densities. Theg values used for these simulations are shown in Table 5.2. Fig 5.4 showsexamples of such simulations. Table 5.3 and Fig. 5.5 give an overview of thedefect densities obtained from the ESR measurements as a function of the

Table 5.2: Principal g values used for the simulation of the defects in the Oimplanted a-SiO2 samples.

Defect g1 g2 g3

POR 2.0017±0.0001 2.0074±0.0001 2.067 [a]NBOHC 2.001±0.0001 2.0095±0.0001 2.078 [a]

OS 2.0022±0.0001 2.0033±0.0001 2.0088±0.0001

[a] g value taken from literature [81].

Table 5.3: The density of the various ESR defects in the O implanted samples.

O implanted E′ NBOHC POR OSsilica (at.%) (1015 mm−3) (1015 mm−3) (1015 mm−3) (1015 mm−3)

0.035 1.7±0.5 6±2 18±5 (N.D.) [a]0.070 0.7±0.2 4±1 17±4 (N.D.) [a]0.350 0.5±0.1 (N.D.)a 32±9 13±40.700 0.08±0.02 (N.D.)a 36±10 42±132.1 0.028±0.008 (N.D.)a 26±8 18±5

[a] N.D. means not detected; estimated below <4×1015 mm−3.


20 G

g=1.99889

O implanted (0.7 at. %)20.5 GHz; 77 K

Si:P

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

g=1.99891Si:P

20 G

O implanted (0.7 at. %)33 GHz; 100 K

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

(a)

(b)

Figure 5.4: ESR spectra of the 0.7 at.% O implanted samples measured in K-band(a) and Q-band (b), exposing the appearance of an additional signal labeled OS.Consistent fitting of the observed spectra (black solid curves) led to decompositioninto two (powder pattern) spectra attributed to the OS (red dotted curves) and thePOR (red solid curves) with the principle g values shown in Table 5.2. The bluedotted curve represents the sum of the two simulations.

5.3 Results 163

0.1 1

0

5

10

15

20

25

30

35

40

45

50 O implanted

E'γ

NBOHC POR OS

defe

ct d

ensi

ty (1

015 m

m-3)

log(at.%)

Figure 5.5: Plot of the density of each component of the measured ESR spectrain the O implanted samples as a function of the implantation concentration.

implanted O concentration. From Table 5.3 it can be seen that for the Oimplanted samples, NBOHCs could only be detected in the samples withthe two smallest concentrations. The PORs, by contrast, are detected inall the samples with changes <15% for the first two concentrations andthen approximately a factor of two increase for the other three concentra-tions. In the spectrum of the sample with an O implantation of 0.35 at.%another defect center, labeled OS, is detected, which density increases byapproximately a factor of 3 for the next concentration, but then decreasesby a factor greater than 2 for the largest concentration.

5.3.2 Silicon implanted samples

Several defects could also be observed in the Si implanted samples includingthe E′γ center, the POR, and a hf doublet with a splitting of 73 G and a zerocrossing g value gc=2.0014±0.0001. Corrections for second order hf effectsbring the g value to g=2.00135±0.0001. A similar doublet has previouslybeen observed in SiO2 thermally grown on Si after VUV irradiation, andwas attributed to O2=Si•–H (see section 1.2.3). The NBOHC is possiblypresent, however, there is no clear experimental evidence likely because ofsensitivity reasons.

In the samples with the larger Si implanted concentrations, an additionalcomponent is observed as a shoulder situated on the low-field flank of the


1.5 at.%

0.5 at.%

0.25 at.%

0.05 at.%

0.025 at.%Si implanted

E'

Si:P

g=1.99889

20.5 GHz; 77 K

10 G

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

E'g=1.99891

Si:P

20 G

1.5 at.%

0.5 at.%

0.25 at.%

0.05 at.%

0.025 at.%

Si implanted~33 GHz; 100 K

dPµ/d

B (a

rb. u

nits

)

magnetic field (G)

Figure 5.6: Overview of measured K and Q-band ESR spectra on fused silica forthe different Si implantation doses. As a measure of dose, the curves are labeledby the calculated at.% of Si atoms implanted in a layer ∼600 nm thick, overwhich the concentration is approximately constant. The size of all samples studiedwas nominally identical. The signal at g=1.99891 stems from a co-mounted Si:Pmarker sample. Signals are scaled to equal marker intensity.

E′ spectrum. The E′ density was determined through double numerical in-tegration of the measured signal. The additional component was includedin these integrations, as the component resolution was considered unreli-able preventing an accurate estimation of this underlying component. Thedetermined E′ densities for the samples with the largest Si implanted con-centrations thus include the density of the additional center, which, there-fore, may be overestimated. An estimate of the upper limit of the defectdensity of the underlying component is ∼2×1016 mm−3. The g value andpeak-to-peak line width of this signal are estimated to be g=2.0026±0.0005and ∆Bpp=(10±5) G, respectively.

An overview of the K and Q-band spectra recorded for the various Si im-planted samples at different implantation concentrations is given in Fig. 5.6.The total number of E′ defects observed in the Si implanted samples withthe largest Si concentrations –see Table 5.4– is one of the largest reportedin literature and, based on the derived E′ densities for the samples withthe two largest Si implantations, we suggest a saturation of the E′ densityfor these samples.

The POR ESR signals had to be simulated to determine the proper de-

5.3 Results 165

Table 5.4: The density of the various ESR defects in the Si implanted samples.

Si implanted E′ 73 G doublet PORsilica (at.%) (1015 mm−3) (1015 mm−3) (1015 mm−3)

0.025 3±1 (N.D.)[a] 15±50.05 2.2±0.7 0.09±0.03 12±40.25 20±6 0.28±0.08 14±40.5 48±15 1.1±0.3 (N.D.)[b]1.5 70±21 1.0±0.3 (N.D.)[b]

[a] N.D. means not detected; estimated below ≤0.07×1015 mm−3.[b] N.D. means not detected; estimated below ≤4×1015 mm−3.

Table 5.5: Principal g values used for the simulation of the POR defect in the Siimplanted samples.

Defect g1 g2 g3

POR 2.0014±0.0001 2.0072±0.0001 2.067[a]

[a] g value taken from literature [81].

fect densities; The principal g values used for these simulations are shownin Table 5.5. Figure 5.7 shows an example of such a simulation. No spec-tral simulations were carried out to determine the E′ spin densities sincewe assume that more accurate results could be obtained from straightfor-ward double numerical integration of the measured component. Indeed,the shape of the E′ component varies with the implanted concentration.Table 5.4 and Fig. 5.8 give an overview of the inferred defect densities as afunction of the Si implantation concentrations.

Since S=1 centers have been reported previously in silica and SiO2 singlecrystals [268, 269, 270, 271] we have made measurements at half the fieldstrength of the ∆MJ=1 transitions, to trace possible ∆MJ=2 transitions.In neither the Si nor the O implanted samples did we detect such transi-tions. On the basis of previous observations [272], the detection limit isestimated at about 1×1012 defects in the sample.


12040 12060 12080 12100 12120 12140

g=1.99891

Si:PdP

µ/dB

(arb

. uni

ts)

Si implanted (0.05 at. %)33 GHz; 100 K

magnetic field (G)

Figure 5.7: Q-band ESR spectrum of the 0.05 at. % Si implanted samples. Thedashed curve represents the simulation of the POR using the principal g valueslisted in Table 5.5

0.1 1

0

20

40

60

log(at.%)

E'γ

73 G doublet POR

defe

ct d

ensi

ty (1

015 m

m-3)

Si implanted

Figure 5.8: Plot of the density of each component of the measured ESR spectrain the Si implanted samples as a function of the implantation concentration.

5.4 Analysis and discussion 167

5.4 Analysis and discussion

5.4.1 The 4.8-4.9 eV optical absorption band

Comparison of optical and ESR data

When comparing the optical absorption data for the 4.8-4.9 eV band withthe ESR date, the E′-type defects can be excluded in this discussion as theseare known to absorb in the 5.7-5.9 range, as discussed later. Before starting,it should be noted that in this comparison of optical band intensities withESR defect densities the relative errors must be considered. In the case ofESR defect densities we estimate that the absolute error is ∼ ±30%. Inthe case of the intensities of the optical bands the error in the measuredoptical densities is ±5%. The error in the intensities of the bands fitted tothe data are more difficult to estimate. Four bands between 4.8 and 4.9 eVwith FWHM between 0.2 and 0.5 eV make the estimate of errors uncertain.Magruder et al. guessed that the errors are ∼ ±50%. The relative errorswere estimated at ±20%. Because of these relatively large errors in theinferred defect densities and band intensities, the comparison of the defectdensities and the optical absorption coefficients is tentative.

The optical data (cf. Fig. 5.1) show that the 4.83 eV absorption has amonotonic increase from the smallest to the largest O concentration. Forthese O implanted samples the POR is detected in all samples but onlyincreases by a factor of ∼2 over the range of concentrations covered in thesesamples, while the 4.83 eV band increases by a factor of ∼5. However, herewe should note that we have additionally detected an ESR signal, labeledOS, in the samples with 0.35 at.% of O and larger. Assuming that, assuggested above, the OS center is due to an O related species, and addingits density to that of theNBOHCs and PORs, the maximum total increasein the added density of all ESR active O related defects obtained whencomparing the 0.035 at.% and the 0.7 at.% samples, is a factor of ∼3.2 (cf.Table 5.3). This factor is still smaller than the maximum increase (factor∼5) obtained optically. Moreover, the total ESR defect density decreasesfor the largest O implantation dose (2.1 at.%), while the optical absorptionstill increases. These findings indicate that in addition to the three ESRactive O related defects (POR, NBOHC, and OS) possibly absorbingin the range 4.8-4.9 eV , at least one additional, probably diamagnetic, Orelated center is formed which contributes to the absorption.

There are various theoretical calculations of the energy states of variousO related defect configurations in silica. Pacchioni and Ierano [273] havecalculated the energies for transitions from the ground state to the first


excited state of a NBOHC (2.2 eV ), of a POR (6.7 eV ), of a bridging O2

(6.8 eV ), and of a hydroxyl ion bonded to a Si (7.6 eV ). In the case of theNBOHC, the authors suggest that there is a possibility of a transition at∼5 eV in agreement with the experimental results. In the case of the otherO related defects the calculated energies are more than 1 eV greater than4.8-4.9 eV . Also, we note that the theoretical link between absorption at4.83 eV and the ESR active POR is qualitative.

Prior work has attributed the optical absorption between 4.8 and 4.9 tothree types of oxygen related defects, NBOHCs, PORs, and O3 molecules[252, 248]. To our knowledge, it has not been suggested that some ofthe absorption could be due to Si related centers. In the Si implantationcase measurable absorption (cf. Fig. 5.1) is generated after the smallestimplanted concentration, but increases only by a factor of 1.6 for the nexttwo concentrations. It increases by a factor of ∼2 in the samples with 0.5at.% and then in the samples with 1.5 at.% the increase is a factor of 15over the absorption in the smallest concentration sample. Based on thisincrease for the largest implanted concentration we propose that there isan incubation concentration for Si related defects, or, possibly, more typesof defects, and that the density of these defects formed increases when thereis a sufficient density of Si ions to form the Si configuration that absorbsin the range 4.8-4.9 eV .

The fact that the absorption between 4.8 and 4.9 eV , after Si implan-tation for the three smallest concentrations, is of approximately the samemagnitude as the optical absorption for all the O samples, the much smallerdensities of the PORs and the lack of detection of NBOHCs for the Siimplanted samples (Table 5.4) indicates that the absorption in all Si im-planted samples is due to different defect centers, presumably Si excessrelated. We suggest that the much larger increase in optical absorption forthe two largest concentrations is due to an increase in the probability offorming the Si configuration which absorbs between 4.8 and 4.9 eV , withincreasing Si concentrations. Based on these observations we suggest thatin the Si implantation case there is at least one Si related defect center ab-sorbing in the 4.8-4.9 eV range. This defect center is probably diamagneticas there is no evidence for a correlated ESR signal. In the Si implantationcase, the ESR data show that the contribution to the absorption by thedetected O related defects (POR) is negligible.


Diamagnetic oxygen related defects

Based on the increase in absorption after the largest O-implantation con-centration compared to the lower O-implantation concentrations and thedecrease in the defect densities of the ESR active O related defects, wesuggest that in addition to these three defects still another O related defectis formed which contributes to the absorption and is probably diamagnetic.Richard et al. [274] have calculated the energy of formation of defects andhave shown that the dominant defect is the oxygen interstitial which formscharged and neutral peroxy bridges. These calculations indicate that thethe probability of forming an oxygen interstitial O−2 is much smaller thanthe probability of forming the peroxy structures. Nishikawa et al. [275]attributed an absorption band observed at 3.8 eV to the neutral peroxybridge (≡Si–O–O–Si≡). If the attribution is correct then such a config-uration cannot contribute to absorption between 4.8 and 4.9 eV . Basedon their discussion it appears that the charge state of the O2 molecule is-2, it is bonded to two Si, and is, therefore, diamagnetic. In the opticalabsorption measurements of the current samples performed by Magruderet al., no absorption band at 3.8 eV was present. However, the assign-ment of the optical absorption band at 3.8 eV to the peroxy bridge hasbeen questioned by Awazu and Kawazoe [252] leaving the possibility thatthe neutral peroxy bridge still absorbs at ∼4.8 eV , and could provide anon-ESR active contribution to the absorption. Another possibility is thatdiamagnetic interstitial O3 molecules account for absorption in the 4.8 eVrange [253].

The OS center

As aforementioned we suggest that the OS center is O related. But whatkind of O related species could account for the observed OS signal? Fromprevious works and compilations [276, 277], a first general observation ap-pears that most O associated defects can be excluded as viable candidatessince they generally have principal g values with at least one g value muchlarger than those inferred for the OS signal. Based on the principal val-ues of the g matrix, for example, it is impossible to assign the OS centerto the interstitial ozonide O−3 ion (g1=2.018, g2=2.011, and g3=2.002) asobserved by Griscom and Friebele [270] in high-potassium content SiO2.Instead, the closest match so far in terms of g values can be found for the’so-called’ O−3 complexes (g1=2.008, g2=2.0045, and g3=2.003) adsorbedon SiO2 surfaces [276]. But, in keeping with scientific rigor, it should be


added that the issue on the identification of the O−3 ion, viz. regardingthe so-called O−3 complexes and anomalous O−3 ions, is not quite satisfac-tory yet [276, 277]. So, we conclude that the atomic source of the OSsignal is unknown. Conclusive assignment would, at least, need additionalinformation on hf structure.

In line with the possible assignment of the diamagnetic O related defectdescribed above, another possible candidate responsible for the OS signalis the positively charged peroxy bridge, perhaps also absorbing at around4.8 eV . According to Richard et al. [274], its formation energy is smallerthan for the neutral peroxy bridge. This complies with the experimentalobservation that the OS center would contribute a (significantly) smallerpart to the 4.8 eV absorption band than the neutral state (cf. Fig. 5.1 andTable 5.3), which is diamagnetic. So, the charged peroxy bridge may alsobe the origin of OS.

Diamagnetic silicon related defects

Based on the comparison of the optical data and ESR data for the Siimplanted samples, we suggest that there is at least one diamagnetic Sirelated defect center that absorbs in the range 4.8-4.9 eV . Next we considersome of the possible Si configurations that could produce such a defect.The calculations of Pacchioni and Ierano [273] give transition energies forthe neutral O3Si–SiO3 defect of 7.5 eV (a dimer), and for the O3Si–Si–SiO3

defect of 6.3 eV (a trimer). Clearly, these energies are much larger than thetransition Magruder et al. observed between 4.8 and 4.9 eV . Since in the Siimplanted samples, the observed change in absorption is small, i.e., a factorof ∼1.6 increase over the first three concentrations, approximately a factorof ∼2.5 increase compared to the smallest concentration for the fourthconcentration, and then a much larger enhancement by a factor of ∼15compared to the smallest concentration for the 1.5 at.% Si concentration,the Si state may be a cluster of Si’s, perhaps, forming a tetragonal structure.

5.4.2 The 5.7-5.9 eV optical absorption band

Comparison of optical and ESR data

To ease comparison, the optical absorption coefficients at 5.85 eV as a func-tion of the log of the implantation concentrations are plotted on the samegraph as the E′ defect densities (see Fig. 5.9). Figure 5.9 (a) shows thatthe optical absorption at 5.85 eV increases with increasing O implantationconcentration while the E′ defect density decreases by a factor of ∼60. This


0.1 10

20

40

60

80

100

120

0.1 1

0

10

20

30

40

50

60

70

80

90

100

absorption at 5.85 eV

E' d

efec

t den

sity

(1015

mm

-3)

Abs

orpt

ion

coef

ficie

nt (1

02 cm

-1)

log(at.%)

[E']

Si implanted

0.1 10

1

2

3

4

5

6

7

8

9

10

11

12

13

0.1 1

0.0

0.5

1.0

1.5

2.0

2.5

absorption at 5.85 eVO implanted

E' d

efec

t den

sity

(1015

mm

-3)

Abs

orpt

ion

coef

ficie

nt (1

02 cm

-1)

log(at.%)

[E']

(a) (b)

Figure 5.9: Absorption coefficient at 5.85 eV and the E′ defect density as afunction of (a) O and (b) Si implantation concentration.

difference is incompatible with the hypothesis that the absorption at 5.85eV would only be due to E′ defects. The current data is, to us, a con-vincing proof of the existence of another optical band or bands between 5.8and 5.9 eV , as suggested previously [256, 257, 248]. Based on the fact thatthe absorption between 5.8 and 5.9 eV increases with increasing O concen-tration we suggest that the additional absorption is due to an O relateddefect. The defect center is probably diamagnetic since no correspondingESR components could be detected.

In the case of the Si implanted samples, as shown in Fig. 5.9 (b), theoptical absorption and the E′γ density increase with increasing implanta-tion dose. The absorption coefficient at 5.85 eV increases approximatelyproportional with the Si concentration –more than a factor of 25 increasebetween the smallest and the largest concentrations. Within the estimatedreasonable errors, we conclude that there is a good qualitative correlation(correlation coefficient R=0.93) between the optical absorption data andthe E′ ESR data for these Si implanted samples. This correlation is con-sistent with the data of Weeks and Sonders [256] and their inference thatthe largest fraction of the absorption at 5.85 eV in their samples was dueto the E′ center.

However, from a different point of view, we emphasize ’reasonable error’noting that accuracies of determining absorption coefficients and defectdensities are reduced when there are overlapping optical bands and ESRcomponents. Along this line, we may dispute the validity of the correlationbetween the E′ densities and the optical absorption coefficient shown in


Fig. 5.9 (b) for the two largest Si concentrations. In fact, we may considerthe E′ data for these two largest concentrations as suggestive for the onsetof E′ density saturation, while the optical data does not indicate such asaturation. If we assume that the differences between the two sets of dataare larger than the estimated errors, the increase in the optical absorptioncannot be due to the E′ centers, leading us to the conclusion that anotherband develops at these concentrations that absorbs between 5.8 and 5.9eV . Since no other ESR active defects, not even S=1 centers, are detectedthat can account for another band, the defect is probably diamagnetic.Furthermore, it is likely that this defect is Si related since it occurs in thesamples with the largest Si implanted concentrations.

Next we might consider some Si configurations that could account forthis additional absorption band. The energy of the transition from theground state to the first excited state of the neutral oxygen vacancy hasbeen calculated as 7.5 eV [273]. If this value is correct then obviously theabsorption of this state will not contribute to the absorption between 5.8and 5.9 eV . With increasing amounts of implanted Si the probability offorming a state in which a two-fold coordinated O2=Si: defect is formedincreases. The charge state of such a site could be 0, +1, or +2. The0 and the +2 states would be diamagnetic. For the 0 charge state thecalculated energy of the first transition is between 5.3 and 5.8 eV [273].Thus this configuration could be the source of the additional absorptionbetween 5.8 and 5.9 eV and would be consistent with the possibility of asecond Si related defect in the 5.7-5.9 eV range that begins to appear athigher concentrations of implanted Si.

E′ density

Saturation of the E′ density as a function of the implantation dose has beenobserved previously. According to Antonini et al. [278] comparing variousdamaging agents, saturation in E′ density is obtained for irradiation ofvitreous SiO2 with 1 MeV protons at the ∼1017 particles/cm2 dose, 46.5MeV nickel ions at ∼1015 particles/cm2, and for 1 MeV electrons at ∼1017

particles/cm2. Bogomolova et al. [279] even reported a decrease in E′

density for different ion implantations at high doses (>6×1015 ions/cm2)in a-SiO2.

In Fig. 5.9 (b) is shown that the density of E′ centers in Si implantedsamples increases with Si dose to the order of 5×1019 cm−3, approximatelyone E′ defect per two implanted Si ions (the maximum Si implantation doseof 1.5 at. % corresponds with the implantation of ∼9×1019 Si atoms per


cm3), which is a very large fraction. However, this density is of the sameorder as the maximum density obtained for 2 MeV proton irradiation [278]and for one case of neutron irradiation [280]. Moritani et al. [281] reportedan E′ density of more than 1020 spins/g, which is rather close to theirevaluated concentration of the displaced oxygen atoms. Weeks reportedthat the E′ defect densities in four different commercial silica samples andone synthetic crystal sample, irradiated with neutrons in a nuclear reactorto doses ranging from 1017 to 3×1020 cm−2, saturated at densities between2×1019 cm−3 and 4×1019 cm−3 [266]. So, the currently observed maximumdefect density of the E′ centers is consistent with previous observations.

The measured shape of the E′ ESR spectra induced by Si ion implan-tation differs from that of the E′ signal observed in γ-irradiated silica [4].These changes become more pronounced for the samples with the largest Siconcentration. We attribute these effects to a larger inhomogeneous broad-ening of the ion implanted samples due to a larger distribution in principalg values, in turn originating from larger distortions of the local environmentof the E′ defects and to increasing dipole-dipole interactions between theE′ spins with increasing densities.

Additional ESR component

As mentioned before (see section 5.3.2), an additional component is ob-served on the low-field flank of the E′ component in the samples withthe largest Si concentrations. A similar center has been observe before,which was assigned to a trivalent Si atom back bonded to two O atomsand one Si atom [279, 201]. However, since no hf interaction originatingfrom this center was ever observed, the assignment should be consideredmerely suggestive. Considering the observed g value and line width, apossible candidate could be S center, tentatively assigned to the E′ likecenters SinO3−n=Si•(n=1,2), likely SiO2=Si•, i.e., the structure proposedin previous works [63, 66] and discussed in section 1.2.2. We note that inthe current case, the resolution of this component is adversely affected byoverlap with the E′ center. In absence of observed hf structure it is notpossible to fully identify this defect.

5.5 Summary and conclusions

The main conclusion of this comparison of the densities of defect centersobserved by ESR and of the optical absorption in the range 5.7-5.9 eV , andespecially in the range 4.8-4.9 eV , is that there are still a lot of unknowns


concerning the origins of the optical absorption. It appears that many para-magnetic as well as diamagnetic defect centers might possibly contributeto the absorption in these ranges. The attribution of an optical absorptionband to one specific defect center (as has been done before [256] for theE′ center and the 5.85 eV absorption) is not straightforward and requiresextreme care. Additional contributions of other defect centers absorbing inthe same range can easily be overlooked due to strong overlap of signals.

In the silica samples implanted with O concentrations ≥0.35 at.%, a sofar unknown ESR component, labeled the OS center, has been resolvedwith principal g values, g1=2.0022, g2=2.0033, and g3=2.0088, as inferredfrom consistent computer simulations of K and Q-band ESR spectra. Asthe defect is not observed in the Si implanted samples, it is likely O re-lated. It may be that this additional O related defect contributes to theabsorption between 4.8 and 4.9 eV . A possible assignment of the OS centercomes from the comparison of the observed principal g values with g valuestaken from literature, pointing towards the direction of the ’so-called’ O−3complex. Another possibility of its atomic origin is the peroxy bridge in theparamagnetic charge state +1. It should be kept in mind, however, thatas long as no hf structure can be observed, these assignments are purelysuggestive and probably one mighty think of many other possible atomicorigins for the OS center.

From the data on the O implanted samples it could be concluded thatbesides the ESR active defects (POR, NBOHC, and OS) that possibly(not certainly) contribute, there should be at least one diamagnetic defectalso contributing to the absorption in the 4.8-4.9 eV range. Combiningthe data with literature, this extra absorption center could possibly beinterstitial ozone molecules or the neutral peroxy bridge.

From the data we could also conclude that beside the E′ center there isanother defect, probably O related, that contributes to the optical absorp-tion between 5.8 and 5.9 eV .

We report for the first time evidence for a band due to a Si relateddefect (or defects) absorbing between 4.8 and 4.9 eV . Based on the depen-dence of the optical absorption of this probably Si related defect on theSi implantation concentration –the formation probability of this Si relatedstate appears implant concentration dependent– we suggest that possiblySi clusters form.

The data on the Si implanted samples show that in these samples, the E′

center is the main defect responsible for the 5.8 eV absorption band. Theoptical absorption between 5.8 and 5.9 eV and the inferred E′ densities cor-


relate within the experimental error for the three smallest concentrations.Based on the trend to saturation of the E′ density at the largest concen-trations of implanted Si and the absence of such a clear trend to saturationin the optical absorption we conclude that another optical band is formedrelated to a configuration of Si ions that absorbs between 5.7 and 5.9 eVand that is probably diamagnetic.

Summary and conclusions

In this work ESR has been used in conjunction with post-manufacture ir-radiation and annealing to observe, monitor and characterize point defectsin the functional oxides SiO2 (nm-sized fumed silica particles and O andSi implanted bulk SiO2) and ZrO2, and in Si/high-κ stacks (P-implanted(100)Si/HfO2 and (100)Si/LaAlO3). This way information could be ob-tained concerning e.g., the specific oxide network structure, the occurrenceof impurity related charge traps, and the nature and stability of the Si/high-κ interface. The main results are summarized below.

The subject of chapter two is the study of point defects in fumed ∼7nm-sized silica nanoparticles using X, K, and Q-band ESR following 10eV irradiation to photodissociate H from passivated defects. Various typesof ESR-active point defects are revealed including the familiar E′ center(generic entity •Si≡O3), EX, the POR, the methyl radical, and an un-known closely axially symmetric center (g‖=2.0041, g⊥=2.0027). Besidesthe PORs, large numbers of other oxygen-hole type defects appear to bepresent also. The exhaustive number of all oxygen-hole centers, includingthe PORs, is determined at ∼0.07 defects/nanoparticle, making this kindof defect highly unlikely as playing a substantial role in narrowing of theoptical band gap, in contrast with previous suggestion.

The main results of this chapter were obtained through monitoring ofthe observed E′ defects as a function of post-formation heating [in vac-uum in the range 850-1115 ◦C or with the sample brought into contactwith ”bulk” Si/SiO2 entities in vacuum at elevated temperatures in therange 1005-1205 ◦C (SiO-vac. anneal)] and treatment (aging, VUV exci-tation), allowing us to assess specific physicochemical structural aspects ofthe nanoparticles. Experimental evidence was presented for the presenceof two different systems of E′ centers. The specific ESR parameters (e.g.,g matrix) of the E′ centers of one batch are are found to be very similar tothose of the E′γ center in bulk fused silica, while the second batch exhibitsa different zero crossing g value and line shape, attributed to variations in

177

178 Summary and conclusions

local structure. It is inferred that the latter E′ system pertains to the outerSiO2 layers, exposing a structural nature different from bulk glassy SiO2.

The SiO-vac. anneals resulted in a drastic increase in E′ defect den-sity with increasing Tan, which enabled us to resolve the primary 29Si hfstructure of the E′ centers located in the core region of the nanoparticles.Detailed analysis of the observed hf spectra reveals several interesting itemspointing to a modification of the specific network structure of the core re-gion of the nanoparticles. An increased hf splitting of 438±2 G is observedcompared to bulk silica (418±2 G) indicating that the core part E′ centersexhibit a more pyramidal defect structure. Moreover, the increased primaryhf splitting indicates that the core of the fumed silica particles is densified,possibly associated with the presence of more low-membered rings in thenm-sized silica network.

In a next chapter the observation by ESR of P-related point defects innm-thick P-implanted HfO2 films on (100)Si after annealing in the range500-900 ◦C and in ZrO2 powder –two oxides prominent in current high-κ insulator research, is reported. Based on the principal g matrices andhf tensors inferred from consistent X, K, and Q-band spectra simulationsand comparison with established P-associated defects in silica, both centersappear similar in nature and are assigned to a P2-type defect –a P substi-tuting a Hf or Zr atom. Both centers were observed in the monoclinicphase of the high-κ oxides with the unpaired electron strongly localizedon the P atom. A sizeable fraction of the incorporated P impurities wasfound to result in ESR active P2-type defects. Within the concern aboutdopant penetration out of Si into the high-κ layers on top, identificationof the dopant-associated defects in the latter appears crucial to which thepresent basic results provide fundamental access; The centers may operateas detrimental charge trapping sites.

The atomic nature of the interface in (100)Si/LaAlO3 structures withnm-thin amorphous LaAlO3 layers of high dielectric constant (κ∼20-27),deposited directly on clean (100)Si by molecular beam deposition at ∼100◦C, is assessed in chapter four through probing of paramagnetic point de-fects. On the as-grown samples K-band ESR indicated the absence of anSi/SiO2-type interface in terms of the archetypal Si-dangling bond-typeSi/SiO2 interface defects (Pb0, Pb1). With no Pb-type defects observed, thisstate is found to persist during subsequent annealing (1 atm N2 or 5 % O2

in N2 ambient) up to Tan∼800 ◦C, referring to a thermally stable abruptSi/LaAlO3 interface, quite in contrast with other high-κ metal oxide/Sistructures. However, in the range Tan∼800-860 ◦C a Si/SiO2-type inter-

Summary and conclusions 179

face starts forming as evidenced by the appearance of Pb0 defects and, withsome delay in Tan, the EX center –an SiO2 associated defect, attestingto significant structural/compositional modification. The peaking of thedefect density vs. Tan curves indicates the SiOx nature of the interlayer tobreak up again upon annealing at Tan≥930 ◦C, possibly related to crystal-lization and/or degrading silicate formation. No specific LaAlO3-specificpoint defects could be traced.

In the last chapter we tried to determine some of the sources of opticalabsorption between 4.8 and 4.9 eV and between 5.8 and 5.9 eV by combin-ing our ESR measurements with the optical absorption measurements madefrom 2.0 to 6.5 eV by Magruder, Weeks, and Weller in two separate seriesof type III silica samples, one implanted with Si and one with O. SeveralESR active defects were observed including the E′γ center, the NBOHC,the POR, a possibly oxygen related center, labeled OS, and a componentaround g=2.0026. As expected, it was found that in the Si case that theE′ center is the main defect responsible for the 5.8 eV band. The trend tosaturation of the E′ density at the largest concentrations of implanted Siand the absence of such a trend in the optical absorption, however, suggestthat another defect center, probably Si-related and diamagnetic, absorbsbetween 5.7 and 5.8 eV . In the O implantation case, the comparison ofthe observed increasing optical absorption at 5.85 eV with the observeddecrease in the E′ defect density with increasing O concentration, indi-cates that an oxygen related band is created and and its intensity increaseswith increasing O concentration. From comparing the changes in absorp-tion between 4.8 and 4.9 eV with the changes in the defect densities ofthe various ESR components, we conclude that there are three ESR-activedefect centers (POR, NBOHC, and OS) that possibly contribute to theoptical absorption between 4.8 and 4.9 eV . The data also indicate the pres-ence of an extra O-related diamagnetic contributor to the absorption in the4.8-4.9 eV range in the O implantation case and a Si-related diamagneticcontributor in the Si implantation case.

Finally, we conclude with some possible perspectives for future research.In this work two high-κ oxides on (100)Si, i.e., HfO2 and LaAlO3 were stud-ied promising for future applications. Recently, the HfO2-based MOSFEThas even been taken into production. However, to meet the projected ad-vancements in semiconductor based technology many other high-κ layerson Si or on other semiconductors (e.g., Ge, GaAs) will need to be tested.More intense ESR work will be needed to characterize, e.g., the nature andstability of the high-κ/semicondictor interface and point defect (intrinsic

180 Summary and conclusions

or impurity related) in the oxide that can act as detrimental charge traps.In the high-κ oxides ZrO2 and HfO2 we succeeded in the observation

and characterization of dopant (P) associated defects. This resulted in aproposition for a possible model for these observed P2-type defect. However,to fully asses the charge trapping behavior of these defects more theoreticalinsight is needed. Hopefully the current experimental breakthrough willencourage theoreticians to start modeling these P2-type defects in high-κoxides.

Not only the impurity (dopant) associated point defects but also the in-trinsic point defects possibly present in the high-κ oxides are of vital impor-tance for the performance and reliability of MOS-based devices. Accordingto theory, the dominant defects in HfO2 are oxygen vacancies exhibitingdefect energy levels in the band gap of the oxide. Up to now, however, noconvincing experimental evidence has been presented in the literature forthe presence of such large concentrations of oxygen vacancies. One of thekey challenges for ESR researchers in the near future will be to providethe undeniable experimental proof for the presence of the much-discussed”oxygen vacancy”.

Bibliography

[1] J. E. Wertz and J. R. Bolton, Electron Spin Resonance: ElementaryTheory and Practical Applications, (McGraw-Hill, New York, 1972).

[2] C. P. Poole, Electron Spin Resonance: A Comprehensive Treatiseon Experimental Techniques, (Interscience Publishers, John Wiley &Sons, New York, 1967).

[3] A. Abragam and B. Bleany, Electron Paramagnetic Resonance ofTransition Ions, (Clarendon Press, Oxford, 1970).

[4] D. L. Griscom, Glass: Science and Technology, volume 4B, editedby D. R. Uhlmann and N. J. Kreidl (Academic Press, N.Y., 1990), p.151.

[5] P. J. Mohr and B. N. Taylor, Rev. Mod. Phys. 77, 64 (2005).

[6] W. Kohn and J. M. Luttinger, Phys. Rev. 97, 883 (1955).

[7] R. G. Barnes and W. V. Smith, Phys. Rev. 93, 95 (1954).

[8] J. R. Morton and K. F. Preston, J. Magn. Reson. 30, 577 (1978).

[9] A. K. Koh and D. J. Miller, At. Data Nucl. Data Tables 33, 235(1985).

[10] C. G. Van de Walle and P. E. Blochl, Phys. Rev. B 47, 4244 (1993).

[11] G. D. Watkins and J. W. Corbett, Phys. Rev. 134, A1359 (1964).

[12] G. Feher, Bell Syst. Techn. J. 36, 449 (1957).

[13] R. Fuller, H. Evans, C. Gamten, B. Czagas, M. Morrison, D. Decrosta,R. Lowry, P. Lenahan, and C. Frye, IEEE Trans. Nuc. Sci. 43, 2565(1996).

181

182

[14] D. L. Griscom, Phys. Rev. B 20, 1823 (1979).

[15] T. E. Tsai and D. L. Griscom, J. Non-Cryst. Sol. 91, 170 (1987).

[16] A. Stesmans, J. Magn. Res. 76, 14 (1988).

[17] A. Stesmans and G. Devos, Phys. Stat. Sol. 110(a), 615 (1988).

[18] G. Breit and I. I. Rabi, Phys. Rev. 38, 2082 (1931).

[19] G. Wilk, R. M. Wallace, and J. M. Anthony, J. Appl. Phys. 89, 5243(2001).

[20] M. Houssa, L. Pantisano, L.-A. Ragnarsson, R. Degraeve, T. Schram,G. Pourtois, S. De Gendt, G. Groeseneken, and M. M. Heyns, Mat.sci. and Eng. R 51, 37 (2006).

[21] International Technology Roadmap for Semiconductors, (Semicon-ductor Industry Association, San Jose, CA, 2005).

[22] S. E. Pratsinis, Prog. Energy Combust. Sci. 24, 197 (1998).

[23] W. J. Stark and S. E. Pratsinis, Powder Technology 126, 103 (2002).

[24] K. L. Brower, Appl. Phys. Lett. 43, 1111 (1983).

[25] G. J. Gerardi, E. H. Poindexter, P. J. Caplan, and N. M. Johnson,Appl. Phys. Lett. 49, 348 (1986).

[26] A. Stesmans and V. V. Afanas’ev, Phys. Rev. B 57, 10030 (1998).

[27] R. Helms and E. H. Poindexter, Rep. Prog. Phys. 57, 791 (1994).

[28] K. L. Brower, Phys. Rev. B 38, 9657 (1988).

[29] A. Stesmans, Phys. Rev. B 48, 2418 (1993).

[30] Y. Nishi, Jpn. J. Appl. Phys. 10, 52 (1971).

[31] P. J. Caplan, E. H. Poindexter, B. E. Deal, and R. R. Razouk, J.Appl. Phys. 50, 5847 (1979).

[32] E. H. Poindexter, P. J. Caplan, B. E. Deal, and R. R. Razouk, J.Appl. Phys. 52, 879 (1981).

[33] A. Stesmans, Appl. Phys. Lett. 48, 972 (1986).

183

[34] A. Stesmans and V. V. Afanas’ev, J. Appl. Phys. 83, 2449 (1998).

[35] A. Stesmans and V. V. Afanas’ev, Appl. Phys. Lett. 72, 2271 (1998).

[36] H. J. von Bardeleben, C. Ortega, A. Grosman, V. Morazzani,J. Siejka, and D. Stievenard, J. Luminescence 57, 301 (1993).

[37] F. C. Rong, J. F. Harvey, E. H. Poindexter, and G. J. Gerardi, Appl.Phys. Lett. 63, 920 (1993).

[38] A. Stesmans, B. Nouwen, and V. V. Afanas’ev, Phys. Rev. B 58,15801 (1998).

[39] A. Stesmans and V. V. Afanas’ev, J. Vac. Sci. Technol. B16, 3108(1998).

[40] W. Futako, M. Mizuochi, and S. Yamasaki, Phys. Rev. Lett. 92,105505 (2004).



[43] A. Stesmans, J. Appl. Phys. 88, 489 (2000).



[46] W. L. Warren, E. H. Poindexter, M. Offenberg, and W. Muller-Warmuth, J. Electrochem. Soc. 139, 872 (1992).

[47] R. H. Silsbee, J. Appl. Phys. 32, 1459 (1961).

[48] D. L. Griscom, Nucl. Instrum. Methods in Phys. Research B 1, 481(1984).

[49] G. Hochstrasser and J. F. Antonini, Surf. Sci. 32, 644 (1972).

[50] R. A. Weeks, J. Appl. Phys. 27, 1376 (1956).

[51] D. L. Griscom, E. J. Friebele, and G. H. Sigel Jr., Solid State Com-mun. 15, 479 (1974).

[52] F. J. Feigl, W. B. Fowler, and K. L. Yip, Solid state Commun. 14,225 (1974).

184

[53] J. K. Rudra and W. B. Fowler, Phys. Rev. B 35, 8223 (1987).

[54] T. Uchino, M. Takahashi, and T. Yoko, Phys. Rev. Lett. 86, 5522(2001).

[55] T. Uchino and T. Yoko, Phys. Rev. B 74, 125203 (2006).

[56] J. F. J. Conley, P. M. Lenahan, H. L. Evans, R. K. Lowry, and T. J.Morthorst, J. Appl. Phys. 76, 2872 (1994).

[57] L. P. Trombetta, G. J. Gerardi, D. J. Di Maria, and E. Tierney, J.Appl. Phys. 64, 2434 (1988).

[58] D. Herve, J. L. Leray, and R. A. B. Devine, J. Appl. Phys. 72, 3634(1992).

[59] V. V. Afanas’ev and S. A., J.Phys.: Condens. Matter 12, 2285 (2000).

[60] E. Holzenkampfer, F. W. Richter, J. Stuke, and U. Voget-Grote, J.Non-Cryst. Solids 32, 327 (1979).

[61] D. L. Griscom, E. J. Friebele, K. J. Long, and J. W. Fleming, J.Appl. Phys. 54, 3743 (1983).


[63] A. Stesmans, B. Nouwen, and V. V. Afanas’ev, Phys. Rev. B 66,045307 (2002).

[64] A. Stesmans, B. Nouwen, D. Pierreux, and V. V. Afanas’ev, Appl.Phys. Lett. 80, 4753 (2002).

[65] A. Stesmans and V. V. Afanas’ev, Microel. Eng. 36, 201 (1997).

[66] A. Stirling and A. Pasquarello, Phys. Rev. B 66, 245201 (2002).


[68] A. Stesmans and F. Scheerlinck, Phys. Rev. B 50, 5204 (1994).

[69] A. Stesmans and F. Scheerlinck, J. Appl. Phys. 75, 1047 (1994).

[70] S. M. Prokes and W. E. Carlos, J. Appl. Phys. 78, 2671 (1995).

[71] W. E. Carlos and S. M. Prokes, J. Appl. Phys. 78, 2129 (1995).

185

[72] A. Baumer, M. Stutzmann, M. S. Brandt, F. C. K. Au, and T. S.Lee, Appl. Phys. Lett. 85, 943 (2004).

[73] R. P. Wang, Appl. Phys. Lett. 88, 142104 (2006).

[74] N. Fukata, J. Chen, T. Sekiguchi, S. Matsushita, O. T., N. Uchida,K. Murakami, T. Tsurui, and S. Ito, Appl. Phys. Lett. 90, 153117(2007).

[75] M. Dohi, H. Yamatani, and T. Fujita, J. Appl. Phys. 91, 815 (2002).

[76] H. J. von Bardeleben, J. L. Cantin, J. J. Ganem, I. Trimaille, andE. P. Gusev, Defects in High-κ Gate Dielectric Stacks, NATO scienceseries, p. 249, edited by E. P. Gusev (Springer, 2006), p. 249.


[78] B. B. Triplett, P. T. Chen, Y. Nishi, P. H. Kasai, J. J. Chambers,and J. Colombo, J. Appl. Phys. 101, 013703 (2007).

[79] H. Kusumoto and M. Shoji, Jpn. J. Appl. Phys. 17, 1678 (1962).

[80] D. L. Griscom, J. Non-Cryst. Solids 31, 241 (1978).

[81] E. J. Friebele, D. L. Griscom, M. A. Stapelbroek, and R. A. Weeks,Phys. Rev. Lett. 42, 1346 (1979).

[82] M. Stapelbroek, D. L. Griscom, E. J. Friebele, and G. H. Sigel Jr., J.Non-Cryst. Solids 32, 313 (1979).

[83] D. L. Griscom and E. J. Friebele, Phys. Rev. B 24, 4896 (1981).

[84] J. Vitko Jr., J. Appl. Phys. 49, 5530 (1978).

[85] Y. J. Uchida, J. Phys. Soc. J. 42, 1937 (1977).

[86] Y. Uchida, J. Isoya, and J. A. Weil, Am. Chem. Soc. 83, 3462 (1979).

[87] P. Lenahan, C. A. Billman, R. Fuller, H. Evance, W. H. Speece,D. Decrosta, and R. Lowry, IEEE Trans. Nuc. Sci. 44, 1834 (1997).

[88] M. Fanciulli, E. Bonera, S. Nokhrin, and G. Pacchioni, Phys. Rev. B74, 134102 (2006).

[89] J. Laegsgaard and K. Stokbro, Phys. Rev. B 61, 12590 (2000).

186

[90] J. Laegsgaard and K. Stokbro, Phys. Rev. B 65, 075208 (2002).

[91] R. A. Weeks and P. J. Bray, J. Chem. Phys. 48, 5 (1968).

[92] W. L. Warren, M. R. Shaneyfelt, D. M. Fleetwood, P. S. Winokur,and S. Montague, IEEE Trans. Nuc. Sci. 42, 1731 (1995).

[93] J. Robertson, Proceedings of Symposium on the Physics and Technol-ogy of Amorphous SiO2, edited by R. A. B. Devine (Plenum Press,N.Y., 1988), p. 91.

[94] M. A. Quevedo-Lopez, M. El-Bouanani, B. E. Gnade, R. M. Wallace,M. R. Visokay, M. Douglas, M. J. Bevan, and L. Colombo, J. Appl.Phys. 92, 3540 (2002).

[95] J. Robertson, Eur. J. Appl. Phys. 28, 265 (2004).

[96] E. P. Gusev, M. Copel, E. Cartier, I. J. R. Baumvol, C. Krug, andM. A. Bribelyuk, Appl. Phys. Lett. 76, 176 (2000).

[97] C. R. Krug, E. B. da Rosa, R. M. de Almeida, J. Morais, I. J.Baumvol, T. D. M. Salgado, and F. C. Stedile, Phys. Rev. Lett.85, 4120 (2000).

[98] P. D. Kirsch, C. S. Kang, J. Lozano, J. C. Lee, and J. G. Ekerdt, J.Appl. Phys. 91, 4354 (2002).

[99] B. Gallas, A. Brunet-Bruneau, S. Fisson, G. Vuye, and J. Rivory, J.Appl. Phys. 92, 1922 (2002).

[100] Y. Nishikawa, T. Yamaguchi, M. Yoshiki, H. Satake, andN. Fukushima, Appl. Phys. Lett. 81, 4386 (2002).

[101] M. R. Visokay, J. J. Chambers, A. L. Rotondaro, A. Shanware, andL. Colombo, Appl. Phys. Lett. 80, 3183 (2002).

[102] M. H. Cho, Y. S. Roh, N. Whang, K. Jeong, H. J. Choi, S. W. Nam,D. H. Ko, J. H. Lee, N. I. Lee, and K. Fujihara, Appl. Phys. Lett.81, 1071 (2002).

[103] M. Copel, M. Bribelyuk, and E. Gusev, Appl. Phys. Lett. 76, 436(2000).

[104] M. Houssa, M. Tuominen, N. Naili, V. V. Afanas’ev, A. Stesmans,S. Haukka, and M. M. Heyns, J. Appl. Phys. 87, 8615 (2000).

187

[105] M. Copel, Appl. Phys. Lett. 82, 1580 (2003).

[106] J. H. Lee and M. Ichikawa, J. Appl. Phys. 92, 1929 (2002).

[107] M. Gutowski, J. E. Jaffe, C. L. Liu, M. Stoker, R. I. Hedge, R. S.Rai, and P. J. Tobin, Appl. Phys. Lett. 80, 1897 (2002).

[108] M. Houssa, A. Stesmans, M. Naili, and M. M. Heyns, Appl. Phys.Lett. 77, 1381 (2000).

[109] J. R. Chavez, R. A. B. Devine, and L. Koltunski, J. Appl. Phys. 90,4284 (2001).

[110] V. V. Afanas’ev and A. Stesmans, Appl. Phys. Lett. 80, 1261 (2002).

[111] R. J. Carter, Intern. Workshop on Gate Insulators, edited by S.Ohmi, F. Fuijta, and H. S. Momose (Jpn. Soc. of Appl. Phys., Tokyo,2001), p. 94.

[112] A. Stesmans and V. V. Afanas‘ev, J. Phys.: Condens. Matter 13,L673 (2001).


[114] S. Baldovino, S. Nokrin, G. Scarel, M. Fanciulli, G. T., and M. S.Brandt, J. Non-Cryst. Solids 322, 168 (2003).

[115] J. L. Cantin and H. J. von Bardeleben, J. Non-Cryst. Solids 303,175 (2002).

[116] B. J. Jones and R. C. Barklie, Microelectron. Eng. 80, 74 (2005).


[118] A. Y. Kang, P. M. Lenahan, J. F. Conley Jr., and R. Solanski, Appl.Phys. Lett. 81, 1128 (2002).

[119] M. Copel and M. C. Reuter, Appl. Phys. Lett. 83, 3398 (2003).

[120] M. L. Green, E. P. Gusev, R. Degraeve, and E. Garfunkel, J. Appl.Phys. 90, 2057 (2001).

[121] L. F. Edge, D. G. Schlom, S. A. Chambers, E. Cicerrella, J. L. Freeouf,B. Hollander, and J. Schubert, Appl. Phys. Lett. 84, 726 (2004).

188

[122] B. H. Lee, L. Kang, R. Nieh, W. J. Qi, and J. C. Lee, Appl. Phys.Lett. 76, 1926 (2000).

[123] M. Houssa, M. Naili, C. Zhao, M. M. Heyns, and A. Stesmans, Semi-cond. Sci. Technol. 16, 31 (2001).

[124] M. J. Torralvo, M. A. Alario, and M. A. Soria, J. Catal. 86, 473(1984).

[125] C. Morterra, E. Giamello, L. Orio, and M. Volante, J. Phys. Chem.94, 3111 (1990).

[126] F. R. Chen, G. Coudurier, J. F. Joly, and J. C. Vedrine, J. Catal.143, 616 (1993).

[127] V. Ramaswamy, B. Tripathi, D. Srinivas, A. V. Ramaswamy, R. Cat-taneo, and R. Prins, J. Catal. 200, 250 (2001).

[128] A. Y. Kang, P. M. Lenahan, and J. F. Conley Jr., Appl. Phys. Lett.83, 3407 (2003).

[129] S. Wright and R. C. Barklie, J. Mater. Sci. in Semicond. Proc. 9, 892(2006).

[130] S. Wright and R. C. Barklie, J. Mat. Sci.: Mat. in Electr. 18, 743(2007).

[131] C. B. Azzoni and A. Paleari, Phys. Rev. B 40, 6518 (1989).

[132] C. B. Azzoni and A. Paleari, Phys. Rev. B 40, 9333 (1989).

[133] J. Matta, J. F. Lamonier, E. Abi-Aad, E. A. Zhilinskaya, andA. Aboukaıs, Phys. Chem. Chem. Phys. 1, 4975 (1999).

[134] Q. Zhao, X. Wang, and T. Cai, Appl. Surf. Sci. 225, 7 (2004).

[135] E. V. Frolova and M. I. Ivanovskaya, Mater. Sci. Eng. C 26, 1106(2006).

[136] V. M. Orera, R. I. Merino, Y. Chen, R. Cases, and P. J. Alonso,Phys. Rev. B 42, 9782 (1990).

[137] H. Liu, L. Feng, X. Zhang, and Q. Xue, J. Phys. Chem. 99, 332(1995).

189

[138] A. Adamski, Z. Sojka, K. Dyrek, and M. Che, Sol. State Ion. 117,113 (1999).

[139] M. G. Cattania, A. Gervasini, F. Morazzoni, R. Scotti, and D. Stru-molo, J. Chem. Soc. Faraday Trans. 1 85, 801 (1989).

[140] A. Martinez-Arias, M. Fernandez-Garcia, C. Belver, J. C. Conesa,and J. Soria, Catal. Lett. 65, 197 (2000).

[141] R. Ben-Michael, D. S. Tannhauser, and J. Genossar, Phys. Rev. B43, 7395 (1991).

[142] M. Occhiuzzi, D. Cordischi, and R. Dragone, J. Phys. Chem. B 106,12464 (2002).

[143] K. K. Ermakovich, V. N. Lazukin, J. V. Chepeleva, and V. I. Alek-sandrov, Sov. Phys. Solid State 18, 1022 (1976).

[144] R. I. Merino, V. M. Orera, E. E. Lomonova, and S. K. Batygov, Phys.Rev. B 52, 6150 (1995).

[145] I. V. Bobricheva, I. A. Stavitsky, V. K. Yermolaev, N. S. Kotsarenko,V. P. Shmachkova, and D. I. Kochubey, Catal. Lett. 56, 23 (1998).

[146] W. Kanzig and M. H. Cohen, Phys. Rev. Lett. 3, 509 (1959).

[147] A. Stesmans, V. V. Afanas’ev, F. Chen, and S. A. Campbell, Appl.Phys. Lett. 84, 4574 (2004).

[148] A. Stesmans, V. V. Afanas’ev, F. Chen, and S. A. Campbell, Mat.Sci. in Semicond. Processing 7, 197 (2004).

[149] J. L. Gole, Z. L. Wang, Z. R. Dai, J. Stout, and M. White, ColloidPolym. sci. 281, 673 (2003).

[150] K. D. Y., G. R. Kowach, D. W. Johnson Jr., S. Bhandarkar, andH. Du, J. Non-Cryst. Solids 342, 18 (2004).

[151] H. K. Kammler and S. E. Pratsinis, Chem. Eng. Process. 39, 219(2000).

[152] M. A. Siddig, S. Radiman, L. S. Jan, and S. V. Muniandy, Colloidsand Surf. A: Pysicochem. Eng. Aspects 254, 215 (2005).

190

[153] T. Uchino, A. Aboshi, S. Kohara, Y. Ohishi, M. Sakashita, andK. Aoki, Phys. Rev. B 69, 155409 (2004).

[154] T. Uchino, A. Sakoh, M. Azuma, M. Takano, M. Takahashi, andT. Yoko, J. Phys.: Condens. Matter 14, 11111 (2002).

[155] S. I. Altman, D. Lee, J. D. Chung, J. Song, and M. Choi, Phys. Rev.B 63, 151402(R) (2001).

[156] A. Roder, W. Kob, and K. Binder, J. Chem. Phys. 114, 7602 (2001).

[157] I. V. Schweigert, K. E. J. Lehtinen, M. J. Carrier, and M. R.Zachariah, Phys. Rev. B 65, 235410 (2002).

[158] A. J. Hurd, D. W. Schaefer, and J. E. Martin, Phys. Rev. A 35, 2361(1987).

[159] T. Campbell, R. K. Kalia, A. Nakano, F. Shimojo, K. Tsuruta,P. Vashishta, and S. Ogata, Phys. Rev. Lett. 82, 4018 (1999).

[160] B. C. Bunker, D. M. Haaland, T. A. Michalske, and W. L. Smith,Surf. Sci. 222, 95 (1989).

[161] H. Saito and T. Hyodo, Mater. Sci. Forum 255-257, 463 (1997).

[162] T. Uchino and T. Yamada, Appl. Phys. Lett. 85, 1164 (2004).

[163] Y. D. Glinka, S. H. Lin, and Y. T. Chen, Phys. Rev. B 66, 035404(2002).

[164] Y. D. Glinka, S. H. Lin, and Y. T. Chen, Appl. Phys. Lett. 75, 778(1999).

[165] C. C. Liu and G. E. Maciel, J. Am. Chem. Soc. 118, 5103 (1996).

[166] S. M. Prokes, W. E. Carlos, L. Seals, S. Lewis, and J. L. Gole, Mater.Lett. 54, 85 (2002).

[167] A. Stesmans and V. V. Afanas’ev, Phys. Rev. B 54, R11129 (1996).

[168] D. L. Griscom, J. Appl. Phys. 58, 2524 (1985).

[169] A. Pusel, U. Wetterauer, and P. Hess, Phys. Rev. Lett. 81, 645(1998).

[170] B. A. Morrow and I. A. Cody, J. Phys. Chem. 80, 1995 (1976).

191

[171] G. J. Young, J. Colloid Sci. 13, 67 (1958).

[172] D. L. Griscom and M. Cook, J. Non-Cryst. Solids 182, 119 (1995).

[173] J. T. Yount, P. M. Lenahan, and P. W. Wyatt, J. Appl. Phys. 77,699 (1995).

[174] R. Tromp, G. W. Rubloff, P. Balk, F. K. Le Goues, and E. J. vanLoenen, Phys. Rev. Lett. 55, 2332 (1985).

[175] Y. Takakuwa, M. Nihei, and M. Miyamoto, Jpn. J. Appl. Phys. Part2 32, L480 (1993).

[176] Y. Takakuwa, M. Nihei, and M. Miyamoto, Appl. Surf. Sci. 117-118,141 (1997).

[177] K. G. Nakamura, H. Kuroki, and M. Kitajima, J. Appl. Phys. 75,4261 (1994).

[178] K. G. Nakamura and M. Kitajima, Appl. Phys. Lett. 65, 2445 (1994).

[179] A. Revesz, J. Electrochem. Soc. 126, 122 (1979).

[180] D. L. Griscom, Nucl. Instrum. Methods in Phys. Research B 46, 12(1990).

[181] P. G. Tello, V. V. Afanas’ev, and A. Stesmans, Microelectron. Eng.72, 81 (2004).


[183] C. L. Marquardt and G. H. Sigel Jr., IEEE Trans. Nucl. Sci. NS-22,2234 (1975).

[184] T. E. Tsai, D. L. Griscom, and E. J. Friebele, Phys. Rev. Lett. 61,444 (1988).

[185] M. E. Zvanut, F. J. Feigl, W. B. Fowler, J. K. Rudra, P. J. Caplan,E. M. Poindexter, and J. D. Zook, Appl. Phys. Lett. 54, 2118 (1989).

[186] B. P. Lemke and D. Haneman, Phys. Rev. B 17, 1893 (1978).

[187] F. J. Feigl and J. H. Anderson, J. Phys. Chem. Solids 31, 575 (1969).

[188] V. A. Radzig, Colloids and Surf. A: Eng. Aspects 74, 91 (1993).

192

[189] V. A. Radzig, Defects in SiO2 and Related Dielectrics: Science andTechnology, NATO Science Series, Series II: Mathematical and Phys-ical Chemistry, edited by G. Paccioni, L. Skuja, and D. L. Griscom(Kluwer Academic Publishers, Dordrecht, 2000), p. 339.

[190] A. A. Bobychev and V. A. Radzig, Kinet. Katal. 29, 551 (1988).

[191] G. K. White, Cryogenics 4, 2 (1964).

[192] M. S. Stoneham, Rev. Mod. Phys. 41, 82 (1969).

[193] S. Agnello, R. Boscaino, G. Buscarino, M. Cannas, and F. M. Gelardi,Phys. Rev. B 66, 113201 (2002).

[194] N. Ishii, M. Kumeda, and T. Shimizu, Jpn. J. Appl. Phys. 20, L673(1981).

[195] Y. Fu and P. A. Fedders, Solid State Commun. 84, 799 (1992).

[196] R. A. B. Devine and J. Arndt, Phys. Rev. B 35, 9376 (1987).

[197] A. H. Edwards and W. B. Fowler, Phys. Rev. B 41, 10816 (1990).

[198] G. Pacchioni and M. Vitiello, Phys. Rev. B 58, 7745 (1998).

[199] R. A. B. Devine and J. Arndt, Phys. Rev. B 42, 2617 (1990).

[200] H. Hosono, H. Kawazoe, K. Oyoshi, and S. Tanaka, J. Non-Cryst.Solids 179, 39 (1994).

[201] M. Mizuguchi, H. Hosono, and H. Kawazoe, Mat. Sci. and Eng. B54, 38 (1998).


[203] W. Gehloff and K. H. Segsa, Phys. Stat. Sol. B 115, 443 (1983).

[204] R. A. B. Devine, R. Dupree, I. Farnan, and J. J. Capponi, Phys. Rev.B 35, 2560 (1987).

[205] Y. T. Thatachari and W. A. Tiller, J. Appl. Phys. 57, 1805 (1985).

[206] L. W. Hobbs and X. Yuan, Defects in SiO2 and Related dielectrics:Science and Technology, NATO science series, Series II: Mathemat-ical and Physical Chemistry, edited by G. Paccioni, L. Skuja, and D.L. Griscom (Kluwer Academic Publishers, Dordrecht, 2000), p. 37.

193

[207] K. Awazu and H. Kawazoe, J. Appl. Phys. 94, 6243 (2003).

[208] Y. Bounouh, L. Chahed, A. Sadki, M. L. Theye, C. Cardinaud,M. Zarrabian, J. von Bardeleben, K. Zellama, and J. L. Fave, Dia-mond Relat. Mater. 4, 492 (1995).

[209] G. K. Walters and T. L. Estle, J. Appl. Phys. 32, 1854 (1961).

[210] M. H. Brodsky and R. S. Title, Phys. Rev. Lett. 23, 581 (1969).

[211] D. L. Griscom, C. I. Merzbacher, R. A. Weeks, and R. A. Zuhr, J.Non-Cryst. Solids 258, 34 (1999).

[212] R. C. Henderson, W. J. Polito, and J. Simpson, Appl. Phys. Lett.16, 15 (1970).

[213] F. Gerson and W. Huber, Electron Spin Resonance Spectroscopy ofOrganic Radicals, (Wiley, Darmstadt, 2003).

[214] W. R. Austin and R. G. Leisure, J. Appl. Phys. 80, 6646 (1996).

[215] C. L. Gardner and E. J. Casey, Can. J. Chem. 46, 207 (1968).

[216] E. J. Friebele, D. L. Griscom, and K. Rau, J. Non-Cryst. Solids 57,167 (1983).

[217] H. M. McConnel, J. Chem. Phys. 25, 709 (1956).

[218] C. Heller, J. Chem. Phys. 36, 175 (1962).

[219] M. A. Quevedo-Lopez, M. El-Bouanani, M. J. Kim, B. E. Gnade,R. M. Wallace, M. R. Visokay, A. Li-Fatou, M. J. Bevan, andL. Colombo, Appl. Phys. Lett. 81, 1609 (2002).

[220] M. A. Quevedo-Lopez, M. R. Visokay, J. Chambers, M. J. Bevan,A. Lifatou, L. Colombo, M. J. Kim, B. E. Gnade, and R. M. Wallace,J. Appl. Phys. 97, 043508 (2005).

[221] K. Suzuki, H. Tashiro, Y. Morisaki, and Y. Sugita, Jpn. J. Appl.Phys. 44, 8286 (2005).

[222] A. Stesmans, K. Clemer, and V. V. Afanas’ev, Phys. Rev. B 72,155335 (2005).

[223] C. Zhao, G. Roebben, M. Heyns, and O. Van Der Biest, Key Eng.Mat. 206-213, 1285 (2002).

194

[224] J. L. Gavartin, A. L. Shluger, A. S. Foster, and G. I. J. Bersuker, J.Appl. Phys. 97, 053704 (2005).

[225] J. Adam and M. D. Rogers, Acta Cryst. 12, 951 (1959).

[226] L. Levien, C. T. Prewitt, and D. J. Weidner, Am. Mineral. 65, 920(1980).

[227] S. Kimura, K. Iwamoto, M. Kadoshima, Y. Nunoshige, A. Ogawa,T. Nabatame, H. Ota, A. Toriumi, and T. Ohishi, in Dielectric Filmsfor Future ULSI Devices -Science and Technology, (JSAP, Kawasaki,2006), p. 126.

[228] K. Xiong, J. Robertson, M. C. Gibson, and S. Clark, J. Appl. Phys.Lett. 87, 183505 (2005).

[229] H. D. B. Gottlob, M. C. Lemme, T. Mollenhauer, T. Wahlbrink, J. K.Efavi, H. Kurz, Y. Stefanov, K. Haberle, R. Komaragiri, T. Ruland,F. Zaunert, and U. Schwalke, J. Non-Cryst. Solids 351, 1558 (2005).

[230] A. Dimoulas, A. Travlos, G. Vellianitis, N. Boukos, and K. Argy-ropoulos, J. Appl. Phys. 90, 4224 (2001).

[231] Z. B. Fang, S. Chen, Y. Y. Zhu, Y. Q. Wu, Y. L. Fan, Y. Y. Wang,and Z. M. Jiang, Nanotechnology 18, 155205 (2007).

[232] G. Scarel, E. Bonera, C. Wiemer, S. Spiga, M. Fanciulli, I. L. Fe-dushkin, H. Schumann, Y. Lebedinskii, and A. Zenkevich, Appl.Phys. Lett. 85, 630 (2004).

[233] L. Edge, D. Schlom, R. Brewer, Y. Chabal, J. Williams, S. Cham-bers, C. Hinkle, G. Lucovsky, Y. Yang, S. Stemmer, M. Copel,B. Hollander, and J. Schubert, Appl. Phys. Lett. 84, 4629 (2004).

[234] B. E. Park and H. Ishiwara, Appl. Phys. Lett. 79, 806 (2001).

[235] B. E. Park and H. Ishiwara, Appl. Phys. Lett. 82, 1197 (2003).

[236] X. B. Lu, X. Zhang, R. Huang, H. B. Lu, Z. H. Chen, W. F. Xiang,M. He, B. L. Cheng, H. W. Zhou, X. P. Wang, C. Z. Wang, and B. Y.Nguyen, Appl. Phys. Lett. 84, 2620 (2004).

[237] Q. Y. Shao, A. D. Li, J. B. Cheng, H. Q. Ling, D. Wu, Z. G. Liu,N. B. Ming, C. Wand, H. W. Zhou, and B. Y. Nguyen, Appl. Phys.A 81, 1181 (2005).

195

[238] P. Sivasubramani, M. J. Kim, B. E. Gnade, R. M. Wallace, L. F.Edge, D. L. Schlom, H. S. Craft, and J. P. Maria, Appl. Phys. Lett.86, 201901 (2005).

[239] J. Lettieri, J. H. Haeni, and D. G. Schlom, J. Vac. Sci. Technol. A20,1332 (2002).

[240] J. I. Pankove, D. E. Carlson, J. E. Berkeyheiser, and R. O. Wance,Phys. Rev. Lett. 51, 2224 (1983).

[241] A. H. Edwards, J. Non-Cryst. Solids 187, 232 (1995).

[242] T. Vondrak and X. Y. Zhu, J. Phys. Chem. B 103, 4892 (1999).

[243] W. Futako, T. Umeda, M. Nishizawa, T. Yasuda, J. Isoya, and S. Ya-masaki, J. Non-Cryst. Solids 299-302, 575 (2002).


[245] H. Hosono, M. Mizuguchi, H. Kawazoe, and T. Ogawa, Appl.Phys.Lett. 74, 2775 (1999).

[246] T. Bakos, S. N. Rashkeev, and S. T. Pantelides, Phys. Rev. B 69,195206 (2004).

[247] Z. Y. Lu, C. J. Nicklaw, D. M. Fleetwood, R. D. Schrimpf, and S. T.Pantelides, Phys. Rev. Lett. 89, 285505 (2002).

[248] L. Skuja, J. Non-Cryst. Solids 239, 16 (1998).

[249] L. N. Skuja, A. N. Strelesky, and A. B. Pakovich, Solid State Com-mun. 50, 1069 (1984).

[250] D. L. Griscom, J. Ceram. Soc. Jpn. 99, 923 (1991).

[251] H. Hosono and R. A. Weeks, J. Non-Cryst. Solids 116, 289 (1990).

[252] K. Awazu and H. Kawazoe, J. Appl. Phys. 68, 3584 (1990).

[253] L. Skuja, M. Hirano, and H. Hosono, Phys. Rev. Lett. 84, 302 (2000).

[254] L. Skuja, M. Mizugichi, H. Hosono, and H. Kawazoe, Nucl. Instrum.Methods Phys. Res. Sect. B 166-167, 711 (2000).

[255] L. Skuja, B. Guttler, D. Schiel, and A. R. Silin, Phys. Rev. B 58,14296 (1998).

196

[256] R. A. Weeks and E. Sonders, Paramagnetic Resonance, volume 2,edited by W. Low (Academic press, N.Y., 1963), p. 869.

[257] R. A. Weeks, R. H. Magruder III, and P. W. Wang, J. Non-Cryst.Solids 149, 122 (1992).

[258] S. Buscarino, G. Agnello and F. M. Gelardi, J. Non-Cryst. Solids351, 1801 (2005).

[259] H. Nishikawa, E. Watanabe, D. Ito, and Y. Ohki, J. Non-Cryst. Solids174, 179 (1994).

[260] R. H. Magruder III, R. A. Weeks, and R. A. Weller, J. Non-Cryst.Solids 322, 58 (2003).

[261] R. H. Magruder III, R. A. Weeks, R. A. Weller, and R. Gaylon, J.Non-Cryst. Solids 345-346, 284 (2004).

[262] R. H. Magruder III, R. A. Weeks, R. A. Weller, and R. Gaylon, J.Non-Cryst. Solids 351, 1727 (2005).

[263] R. H. Magruder III, A. Stesmans, K. Clemer, R. A. Weeks, and R. A.Weller, J. Non-Cryst. Solids 352, 3027 (2006).

[264] R. H. Magruder III, A. Stesmans, K. Clemer, R. A. Weeks, and R. A.Weller, J. Appl. Phys. 100, 033517 (2006).

[265] J. G. Castle and D. W. Feldman, J. Appl. Phys. 36, 124 (1965).

[266] R. A. Weeks, J. Non-Cryst. Solids 179, 1 (1994).

[267] T. Purcell and R. A. Weeks, Phys. Chem. Glasses 10, 198 (1969).

[268] R. B. Bossoli, M. G. Jani, and L. E. Halliburton, Solid State Com-mun. 44, 213 (1982).

[269] R. A. Weeks and M. Abraham, Bull. Am. Phys. Soc. 10, 374 (1965).

[270] D. L. Griscom and J. L. Friebele, Phys. Rev. B 34, 7524 (1986).

[271] R. Tohmon, Y. Shimogaichi, Y. Tsuta, S. Munekuni, Y. Ohki,Y. Hama, and K. Nagasaka, Phys. Rev. B 41, 7258 (1990).

[272] A. Stesmans, B. Nouwen, and K. Iakoubovskii, J. Phys.: Condens.Matter 12, 7807 (2000).

197

[273] G. Pacchioni and G. Ierano, Phys. Rev. B 57, 818 (1998).

[274] N. Richard, L. Martin-Samos, G. Roma, Y. Limoge, and J. Crocom-bette, J. Non.-Cryst. Solids 351, 1825 (2005).

[275] H. Nishikawa, R. Tohmon, K. Nagasawa, and Y. Hama, J. Appl.Phys. 65, 4672 (1989).

[276] M. Che and A. J. Tench, Adv. Catal. 32, 1 (1983).

[277] L. D. Bogomolova, V. A. Jachkin, S. A. Prushinski, S. A. Dmitriev,S. V. Stefanovsky, Y. G. Teplyakov, F. Caccavale, E. Cattaruzza,R. Bertoncello, and F. Trivillin, J. Non-Cryst. Solids 210, 101 (1997).

[278] M. Antonini, P. Camagni, P. N. Gibson, and A. Manara, Radiat. Eff.65, 41 (1982).

[279] L. D. Bogomolova, Y. G. Telpyakov, A. A. Deshkovskaya, and F. Cac-cavale, J. Non-Cryst. Solids 202, 185 (1995).

[280] D. K. Stevens, W. J. Stur, and R. H. Silsbee, J. Appl. Phys. 29, 66(1958).

[281] K. Moritani, Y. Teraoko, I. Takagi, and H. Mariyama, J. Nucl. Mater.329-333, 988 (2004).

List of Publications

1. A. Stesmans, V. V. Afanas’ev, K. Clemer, F. Chen, and S. A. Camp-bell,Point defects and traps in stacks of ultra thin high-κ metal oxides onSi probed by electron spin resonance: the Si/HfO2 system and N in-corporation,Electrochemical Society Proceedings Volume 2005-01.

2. A. Stesmans, K. Clemer, and V. V. Afanas’ev,Electron spin resonance probing of fundamental point defects in nanometer-sized silica particles,Phys. Rev. B 72, 155335 (2005).

3. A. Stesmans, K. Clemer, and V. V. Afanas’ev,Are intrinsic point defects inadequate as origin of optical band gapnarrowing in fumed silica nanoparticles?,J. Phys.: Condens. Matter 17, L393 (2005).

4. A. Stesmans, K. Clemer, and V. V. Afanas’ev,Electron spin resonance probing of fundamental point defects in nm-sized silica particles,J. Non-Cryst. Solids 351, 1764 (2005).

5. A. Stesmans, K. Clemer, and V. V. Afanas’ev,Electron spin probing of E′-type defects in fumed silica nanoparticles,Mat. Sci. and Eng. C 26, 766 (2006).

6. R. H. Magruder, A. Stesmans, K. Clemer, R. A. Weeks, and R. A.Weller,Origins of optical absorption between 4.8 and 4.9 eV in silica im-planted with Si or O,J. Non-Cryst. Solids 352, 3027 (2006).

199

200

7. R. H. Magruder, A. Stesmans, K. Clemer, R. A. Weeks, and R. A.Weller,Sources of optical absorption between 5.7 and 5.9 eV in silica im-planted with Si or O,J. Appl. Phys. 100, 033517 (2006).

8. A. Stesmans, K. Clemer, V. V. Afanas’ev, L. F. Edge, and D. G.Schlom,Nature and stability of the (100)Si/LaAlO3 interface probed by para-magnetic defects,Appl. Phys. Lett. 89, 112121 (2006).

9. K. Beerten, J. Lomax, K. Clemer, A. Stesmans, and U. Radtke,On the use of Ti centers for estimating burial ages of Pleistocenesedimentary quartz: Multiple-grain data from Australia,Quaternary Geochronology 1, 151 (2006).

10. A. Stesmans, V. V. Afanas’ev and K. Clemer,Probing point defects in stacks of ultrathin high-κ metal oxides onsemiconductors by electron spin resonance: the Si/HfO2 vs the Ge/HfO2

system,Electrochemical Society Proceedings Volume 2006-01.

11. K. Clemer, A. Stesmans, and V. V. Afanas’ev,Observation of a P-associated defect in HfO2 nanolayers on (100)Siby electron spin resonance,Appl. Phys. Lett. 90, 142116 (2007).

12. K. Clemer, A. Stesmans, and V. V. Afanas’ev, L. F. Edge, and D. G.Schlom,Analysis of the (100)Si/LaAlO3 structure by electron spin resonance:nature of the interface,J. Mat. Sci. and Eng. 18, 735 (2007).

13. A. Stesmans, K. Clemer, and V. V. Afanas’ev,Electron spin resonance observation of a P-associated defect in HfO2

films on (100)Si,Microel. Eng. 84, 2358 (2007).

14. K. Clemer, A. Stesmans, and V. V. Afanas’ev, L. F. Edge, and D. G.Schlom,Paramagnetic point defect in (100)Si/LaAlO3 structures: nature and

201

stability of the interface,J. Appl. Phys. 102, 034516 (2007).

15. K. Clemer, A. Stesmans, and V. V. Afanas’ev,Paramagnetic intrinsic point defects in nm-sized silica particles: In-teraction with SiO at elevated temperatures,Mat. Sci. and Eng. C 27, 1475 (2007).

16. A. Stesmans, K. Clemer, P. Somers, V. V. Afanas’ev,Probing semiconductor/insulator heterostructures through electron spinresonance of point defects: Interfaces, interlayers, and stress,Material Research Society Proceedings Fall meeting 2006 (in print).

17. A. Stesmans, K. Clemer, V. V. Afanas’ev,The E′ center as a probe of structural properties of nanometer-sizedsilica particles,J. Non-Cryst. Solids (submitted).

18. K. Clemer, A. Stesmans, and V. V. Afanas’ev,P-associated defects in the high-κ HfO2 and ZrO2 insulators revealedby electron spin resonance,Phys. Rev. B (submitted).

19. K. Clemer, A. Stesmans, and V. V. Afanas’ev,The primary 29Si hyperfine structure of E′ centers in nm-sized silica:probing of the microscopic network structure,Phys. Rev. B (submitted).

c© 2007 Faculteit Wetenschappen, Geel Huis, Kasteelpark Arenberg 11,3001 Heverlee (Leuven)

Alle rechten voorbehouden. Niets uit deze uitgave mag worden vermenig-vuldigd en/of openbaar gemaakt worden door middel van druk, fotokopie,microfilm, elektronisch of op welke andere wijze ook zonder voorafgaande-lijke schriftelijke toestemming van de uitgever.

All rights reserved. No part of the publication may be reproduced in anyform by print, photoprint, microfilm, electronic or any other means withoutwritten permission from the publisher.

ISBN 978-90-8649-124-7D/2007/10.705/46

Page 1 of 1 - core.ac.uk · Elektron spin resonantie (ESR) is hiervoor de uitgelezen techniek....

Documents

Transcript of Page 1 of 1 - core.ac.uk · Elektron spin resonantie (ESR) is hiervoor de uitgelezen techniek....